]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
0d905bca | 42 | #include <linux/perf_counter.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
5517d86b | 67 | #include <linux/reciprocal_div.h> |
dff06c15 | 68 | #include <linux/unistd.h> |
f5ff8422 | 69 | #include <linux/pagemap.h> |
8f4d37ec | 70 | #include <linux/hrtimer.h> |
30914a58 | 71 | #include <linux/tick.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 GH |
79 | #include "sched_cpupri.h" |
80 | ||
a8d154b0 | 81 | #define CREATE_TRACE_POINTS |
ad8d75ff | 82 | #include <trace/events/sched.h> |
a8d154b0 | 83 | |
1da177e4 LT |
84 | /* |
85 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
86 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
87 | * and back. | |
88 | */ | |
89 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
90 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
91 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
92 | ||
93 | /* | |
94 | * 'User priority' is the nice value converted to something we | |
95 | * can work with better when scaling various scheduler parameters, | |
96 | * it's a [ 0 ... 39 ] range. | |
97 | */ | |
98 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
99 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
100 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
101 | ||
102 | /* | |
d7876a08 | 103 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 104 | */ |
d6322faf | 105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 106 | |
6aa645ea IM |
107 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
108 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
109 | ||
1da177e4 LT |
110 | /* |
111 | * These are the 'tuning knobs' of the scheduler: | |
112 | * | |
a4ec24b4 | 113 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
114 | * Timeslices get refilled after they expire. |
115 | */ | |
1da177e4 | 116 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 117 | |
d0b27fa7 PZ |
118 | /* |
119 | * single value that denotes runtime == period, ie unlimited time. | |
120 | */ | |
121 | #define RUNTIME_INF ((u64)~0ULL) | |
122 | ||
5517d86b | 123 | #ifdef CONFIG_SMP |
fd2ab30b SN |
124 | |
125 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
126 | ||
5517d86b ED |
127 | /* |
128 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
129 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
130 | */ | |
131 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
132 | { | |
133 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
134 | } | |
135 | ||
136 | /* | |
137 | * Each time a sched group cpu_power is changed, | |
138 | * we must compute its reciprocal value | |
139 | */ | |
140 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
141 | { | |
142 | sg->__cpu_power += val; | |
143 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
144 | } | |
145 | #endif | |
146 | ||
e05606d3 IM |
147 | static inline int rt_policy(int policy) |
148 | { | |
3f33a7ce | 149 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
150 | return 1; |
151 | return 0; | |
152 | } | |
153 | ||
154 | static inline int task_has_rt_policy(struct task_struct *p) | |
155 | { | |
156 | return rt_policy(p->policy); | |
157 | } | |
158 | ||
1da177e4 | 159 | /* |
6aa645ea | 160 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 161 | */ |
6aa645ea IM |
162 | struct rt_prio_array { |
163 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
164 | struct list_head queue[MAX_RT_PRIO]; | |
165 | }; | |
166 | ||
d0b27fa7 | 167 | struct rt_bandwidth { |
ea736ed5 IM |
168 | /* nests inside the rq lock: */ |
169 | spinlock_t rt_runtime_lock; | |
170 | ktime_t rt_period; | |
171 | u64 rt_runtime; | |
172 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
173 | }; |
174 | ||
175 | static struct rt_bandwidth def_rt_bandwidth; | |
176 | ||
177 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
178 | ||
179 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
180 | { | |
181 | struct rt_bandwidth *rt_b = | |
182 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
183 | ktime_t now; | |
184 | int overrun; | |
185 | int idle = 0; | |
186 | ||
187 | for (;;) { | |
188 | now = hrtimer_cb_get_time(timer); | |
189 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
190 | ||
191 | if (!overrun) | |
192 | break; | |
193 | ||
194 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
195 | } | |
196 | ||
197 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
198 | } | |
199 | ||
200 | static | |
201 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
202 | { | |
203 | rt_b->rt_period = ns_to_ktime(period); | |
204 | rt_b->rt_runtime = runtime; | |
205 | ||
ac086bc2 PZ |
206 | spin_lock_init(&rt_b->rt_runtime_lock); |
207 | ||
d0b27fa7 PZ |
208 | hrtimer_init(&rt_b->rt_period_timer, |
209 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
210 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
211 | } |
212 | ||
c8bfff6d KH |
213 | static inline int rt_bandwidth_enabled(void) |
214 | { | |
215 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
216 | } |
217 | ||
218 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
219 | { | |
220 | ktime_t now; | |
221 | ||
cac64d00 | 222 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
223 | return; |
224 | ||
225 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
226 | return; | |
227 | ||
228 | spin_lock(&rt_b->rt_runtime_lock); | |
229 | for (;;) { | |
7f1e2ca9 PZ |
230 | unsigned long delta; |
231 | ktime_t soft, hard; | |
232 | ||
d0b27fa7 PZ |
233 | if (hrtimer_active(&rt_b->rt_period_timer)) |
234 | break; | |
235 | ||
236 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
237 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
238 | |
239 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
240 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
241 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
242 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 243 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 PZ |
244 | } |
245 | spin_unlock(&rt_b->rt_runtime_lock); | |
246 | } | |
247 | ||
248 | #ifdef CONFIG_RT_GROUP_SCHED | |
249 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
250 | { | |
251 | hrtimer_cancel(&rt_b->rt_period_timer); | |
252 | } | |
253 | #endif | |
254 | ||
712555ee HC |
255 | /* |
256 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
257 | * detach_destroy_domains and partition_sched_domains. | |
258 | */ | |
259 | static DEFINE_MUTEX(sched_domains_mutex); | |
260 | ||
052f1dc7 | 261 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 262 | |
68318b8e SV |
263 | #include <linux/cgroup.h> |
264 | ||
29f59db3 SV |
265 | struct cfs_rq; |
266 | ||
6f505b16 PZ |
267 | static LIST_HEAD(task_groups); |
268 | ||
29f59db3 | 269 | /* task group related information */ |
4cf86d77 | 270 | struct task_group { |
052f1dc7 | 271 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
272 | struct cgroup_subsys_state css; |
273 | #endif | |
052f1dc7 | 274 | |
6c415b92 AB |
275 | #ifdef CONFIG_USER_SCHED |
276 | uid_t uid; | |
277 | #endif | |
278 | ||
052f1dc7 | 279 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
280 | /* schedulable entities of this group on each cpu */ |
281 | struct sched_entity **se; | |
282 | /* runqueue "owned" by this group on each cpu */ | |
283 | struct cfs_rq **cfs_rq; | |
284 | unsigned long shares; | |
052f1dc7 PZ |
285 | #endif |
286 | ||
287 | #ifdef CONFIG_RT_GROUP_SCHED | |
288 | struct sched_rt_entity **rt_se; | |
289 | struct rt_rq **rt_rq; | |
290 | ||
d0b27fa7 | 291 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 292 | #endif |
6b2d7700 | 293 | |
ae8393e5 | 294 | struct rcu_head rcu; |
6f505b16 | 295 | struct list_head list; |
f473aa5e PZ |
296 | |
297 | struct task_group *parent; | |
298 | struct list_head siblings; | |
299 | struct list_head children; | |
29f59db3 SV |
300 | }; |
301 | ||
354d60c2 | 302 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 303 | |
6c415b92 AB |
304 | /* Helper function to pass uid information to create_sched_user() */ |
305 | void set_tg_uid(struct user_struct *user) | |
306 | { | |
307 | user->tg->uid = user->uid; | |
308 | } | |
309 | ||
eff766a6 PZ |
310 | /* |
311 | * Root task group. | |
312 | * Every UID task group (including init_task_group aka UID-0) will | |
313 | * be a child to this group. | |
314 | */ | |
315 | struct task_group root_task_group; | |
316 | ||
052f1dc7 | 317 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
318 | /* Default task group's sched entity on each cpu */ |
319 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
320 | /* Default task group's cfs_rq on each cpu */ | |
321 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 322 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
323 | |
324 | #ifdef CONFIG_RT_GROUP_SCHED | |
325 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
326 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 327 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 328 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 329 | #define root_task_group init_task_group |
9a7e0b18 | 330 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 331 | |
8ed36996 | 332 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
333 | * a task group's cpu shares. |
334 | */ | |
8ed36996 | 335 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 336 | |
57310a98 PZ |
337 | #ifdef CONFIG_SMP |
338 | static int root_task_group_empty(void) | |
339 | { | |
340 | return list_empty(&root_task_group.children); | |
341 | } | |
342 | #endif | |
343 | ||
052f1dc7 | 344 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
345 | #ifdef CONFIG_USER_SCHED |
346 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 347 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 348 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 349 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 350 | |
cb4ad1ff | 351 | /* |
2e084786 LJ |
352 | * A weight of 0 or 1 can cause arithmetics problems. |
353 | * A weight of a cfs_rq is the sum of weights of which entities | |
354 | * are queued on this cfs_rq, so a weight of a entity should not be | |
355 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
356 | * (The default weight is 1024 - so there's no practical |
357 | * limitation from this.) | |
358 | */ | |
18d95a28 | 359 | #define MIN_SHARES 2 |
2e084786 | 360 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 361 | |
052f1dc7 PZ |
362 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
363 | #endif | |
364 | ||
29f59db3 | 365 | /* Default task group. |
3a252015 | 366 | * Every task in system belong to this group at bootup. |
29f59db3 | 367 | */ |
434d53b0 | 368 | struct task_group init_task_group; |
29f59db3 SV |
369 | |
370 | /* return group to which a task belongs */ | |
4cf86d77 | 371 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 372 | { |
4cf86d77 | 373 | struct task_group *tg; |
9b5b7751 | 374 | |
052f1dc7 | 375 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
376 | rcu_read_lock(); |
377 | tg = __task_cred(p)->user->tg; | |
378 | rcu_read_unlock(); | |
052f1dc7 | 379 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
380 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
381 | struct task_group, css); | |
24e377a8 | 382 | #else |
41a2d6cf | 383 | tg = &init_task_group; |
24e377a8 | 384 | #endif |
9b5b7751 | 385 | return tg; |
29f59db3 SV |
386 | } |
387 | ||
388 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 389 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 390 | { |
052f1dc7 | 391 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
392 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
393 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 394 | #endif |
6f505b16 | 395 | |
052f1dc7 | 396 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
397 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
398 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 399 | #endif |
29f59db3 SV |
400 | } |
401 | ||
402 | #else | |
403 | ||
57310a98 PZ |
404 | #ifdef CONFIG_SMP |
405 | static int root_task_group_empty(void) | |
406 | { | |
407 | return 1; | |
408 | } | |
409 | #endif | |
410 | ||
6f505b16 | 411 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
412 | static inline struct task_group *task_group(struct task_struct *p) |
413 | { | |
414 | return NULL; | |
415 | } | |
29f59db3 | 416 | |
052f1dc7 | 417 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 418 | |
6aa645ea IM |
419 | /* CFS-related fields in a runqueue */ |
420 | struct cfs_rq { | |
421 | struct load_weight load; | |
422 | unsigned long nr_running; | |
423 | ||
6aa645ea | 424 | u64 exec_clock; |
e9acbff6 | 425 | u64 min_vruntime; |
6aa645ea IM |
426 | |
427 | struct rb_root tasks_timeline; | |
428 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
429 | |
430 | struct list_head tasks; | |
431 | struct list_head *balance_iterator; | |
432 | ||
433 | /* | |
434 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
435 | * It is set to NULL otherwise (i.e when none are currently running). |
436 | */ | |
4793241b | 437 | struct sched_entity *curr, *next, *last; |
ddc97297 | 438 | |
5ac5c4d6 | 439 | unsigned int nr_spread_over; |
ddc97297 | 440 | |
62160e3f | 441 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
442 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
443 | ||
41a2d6cf IM |
444 | /* |
445 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
446 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
447 | * (like users, containers etc.) | |
448 | * | |
449 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
450 | * list is used during load balance. | |
451 | */ | |
41a2d6cf IM |
452 | struct list_head leaf_cfs_rq_list; |
453 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
454 | |
455 | #ifdef CONFIG_SMP | |
c09595f6 | 456 | /* |
c8cba857 | 457 | * the part of load.weight contributed by tasks |
c09595f6 | 458 | */ |
c8cba857 | 459 | unsigned long task_weight; |
c09595f6 | 460 | |
c8cba857 PZ |
461 | /* |
462 | * h_load = weight * f(tg) | |
463 | * | |
464 | * Where f(tg) is the recursive weight fraction assigned to | |
465 | * this group. | |
466 | */ | |
467 | unsigned long h_load; | |
c09595f6 | 468 | |
c8cba857 PZ |
469 | /* |
470 | * this cpu's part of tg->shares | |
471 | */ | |
472 | unsigned long shares; | |
f1d239f7 PZ |
473 | |
474 | /* | |
475 | * load.weight at the time we set shares | |
476 | */ | |
477 | unsigned long rq_weight; | |
c09595f6 | 478 | #endif |
6aa645ea IM |
479 | #endif |
480 | }; | |
1da177e4 | 481 | |
6aa645ea IM |
482 | /* Real-Time classes' related field in a runqueue: */ |
483 | struct rt_rq { | |
484 | struct rt_prio_array active; | |
63489e45 | 485 | unsigned long rt_nr_running; |
052f1dc7 | 486 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
487 | struct { |
488 | int curr; /* highest queued rt task prio */ | |
398a153b | 489 | #ifdef CONFIG_SMP |
e864c499 | 490 | int next; /* next highest */ |
398a153b | 491 | #endif |
e864c499 | 492 | } highest_prio; |
6f505b16 | 493 | #endif |
fa85ae24 | 494 | #ifdef CONFIG_SMP |
73fe6aae | 495 | unsigned long rt_nr_migratory; |
a1ba4d8b | 496 | unsigned long rt_nr_total; |
a22d7fc1 | 497 | int overloaded; |
917b627d | 498 | struct plist_head pushable_tasks; |
fa85ae24 | 499 | #endif |
6f505b16 | 500 | int rt_throttled; |
fa85ae24 | 501 | u64 rt_time; |
ac086bc2 | 502 | u64 rt_runtime; |
ea736ed5 | 503 | /* Nests inside the rq lock: */ |
ac086bc2 | 504 | spinlock_t rt_runtime_lock; |
6f505b16 | 505 | |
052f1dc7 | 506 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
507 | unsigned long rt_nr_boosted; |
508 | ||
6f505b16 PZ |
509 | struct rq *rq; |
510 | struct list_head leaf_rt_rq_list; | |
511 | struct task_group *tg; | |
512 | struct sched_rt_entity *rt_se; | |
513 | #endif | |
6aa645ea IM |
514 | }; |
515 | ||
57d885fe GH |
516 | #ifdef CONFIG_SMP |
517 | ||
518 | /* | |
519 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
520 | * variables. Each exclusive cpuset essentially defines an island domain by |
521 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
522 | * exclusive cpuset is created, we also create and attach a new root-domain |
523 | * object. | |
524 | * | |
57d885fe GH |
525 | */ |
526 | struct root_domain { | |
527 | atomic_t refcount; | |
c6c4927b RR |
528 | cpumask_var_t span; |
529 | cpumask_var_t online; | |
637f5085 | 530 | |
0eab9146 | 531 | /* |
637f5085 GH |
532 | * The "RT overload" flag: it gets set if a CPU has more than |
533 | * one runnable RT task. | |
534 | */ | |
c6c4927b | 535 | cpumask_var_t rto_mask; |
0eab9146 | 536 | atomic_t rto_count; |
6e0534f2 GH |
537 | #ifdef CONFIG_SMP |
538 | struct cpupri cpupri; | |
539 | #endif | |
7a09b1a2 VS |
540 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
541 | /* | |
542 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
543 | * used when most cpus are idle in the system indicating overall very | |
544 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
545 | */ | |
546 | unsigned int sched_mc_preferred_wakeup_cpu; | |
547 | #endif | |
57d885fe GH |
548 | }; |
549 | ||
dc938520 GH |
550 | /* |
551 | * By default the system creates a single root-domain with all cpus as | |
552 | * members (mimicking the global state we have today). | |
553 | */ | |
57d885fe GH |
554 | static struct root_domain def_root_domain; |
555 | ||
556 | #endif | |
557 | ||
1da177e4 LT |
558 | /* |
559 | * This is the main, per-CPU runqueue data structure. | |
560 | * | |
561 | * Locking rule: those places that want to lock multiple runqueues | |
562 | * (such as the load balancing or the thread migration code), lock | |
563 | * acquire operations must be ordered by ascending &runqueue. | |
564 | */ | |
70b97a7f | 565 | struct rq { |
d8016491 IM |
566 | /* runqueue lock: */ |
567 | spinlock_t lock; | |
1da177e4 LT |
568 | |
569 | /* | |
570 | * nr_running and cpu_load should be in the same cacheline because | |
571 | * remote CPUs use both these fields when doing load calculation. | |
572 | */ | |
573 | unsigned long nr_running; | |
6aa645ea IM |
574 | #define CPU_LOAD_IDX_MAX 5 |
575 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 576 | #ifdef CONFIG_NO_HZ |
15934a37 | 577 | unsigned long last_tick_seen; |
46cb4b7c SS |
578 | unsigned char in_nohz_recently; |
579 | #endif | |
d8016491 IM |
580 | /* capture load from *all* tasks on this cpu: */ |
581 | struct load_weight load; | |
6aa645ea IM |
582 | unsigned long nr_load_updates; |
583 | u64 nr_switches; | |
23a185ca | 584 | u64 nr_migrations_in; |
6aa645ea IM |
585 | |
586 | struct cfs_rq cfs; | |
6f505b16 | 587 | struct rt_rq rt; |
6f505b16 | 588 | |
6aa645ea | 589 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
590 | /* list of leaf cfs_rq on this cpu: */ |
591 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
592 | #endif |
593 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 594 | struct list_head leaf_rt_rq_list; |
1da177e4 | 595 | #endif |
1da177e4 LT |
596 | |
597 | /* | |
598 | * This is part of a global counter where only the total sum | |
599 | * over all CPUs matters. A task can increase this counter on | |
600 | * one CPU and if it got migrated afterwards it may decrease | |
601 | * it on another CPU. Always updated under the runqueue lock: | |
602 | */ | |
603 | unsigned long nr_uninterruptible; | |
604 | ||
36c8b586 | 605 | struct task_struct *curr, *idle; |
c9819f45 | 606 | unsigned long next_balance; |
1da177e4 | 607 | struct mm_struct *prev_mm; |
6aa645ea | 608 | |
3e51f33f | 609 | u64 clock; |
6aa645ea | 610 | |
1da177e4 LT |
611 | atomic_t nr_iowait; |
612 | ||
613 | #ifdef CONFIG_SMP | |
0eab9146 | 614 | struct root_domain *rd; |
1da177e4 LT |
615 | struct sched_domain *sd; |
616 | ||
a0a522ce | 617 | unsigned char idle_at_tick; |
1da177e4 | 618 | /* For active balancing */ |
3f029d3c | 619 | int post_schedule; |
1da177e4 LT |
620 | int active_balance; |
621 | int push_cpu; | |
d8016491 IM |
622 | /* cpu of this runqueue: */ |
623 | int cpu; | |
1f11eb6a | 624 | int online; |
1da177e4 | 625 | |
a8a51d5e | 626 | unsigned long avg_load_per_task; |
1da177e4 | 627 | |
36c8b586 | 628 | struct task_struct *migration_thread; |
1da177e4 LT |
629 | struct list_head migration_queue; |
630 | #endif | |
631 | ||
dce48a84 TG |
632 | /* calc_load related fields */ |
633 | unsigned long calc_load_update; | |
634 | long calc_load_active; | |
635 | ||
8f4d37ec | 636 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
637 | #ifdef CONFIG_SMP |
638 | int hrtick_csd_pending; | |
639 | struct call_single_data hrtick_csd; | |
640 | #endif | |
8f4d37ec PZ |
641 | struct hrtimer hrtick_timer; |
642 | #endif | |
643 | ||
1da177e4 LT |
644 | #ifdef CONFIG_SCHEDSTATS |
645 | /* latency stats */ | |
646 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
647 | unsigned long long rq_cpu_time; |
648 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
649 | |
650 | /* sys_sched_yield() stats */ | |
480b9434 | 651 | unsigned int yld_count; |
1da177e4 LT |
652 | |
653 | /* schedule() stats */ | |
480b9434 KC |
654 | unsigned int sched_switch; |
655 | unsigned int sched_count; | |
656 | unsigned int sched_goidle; | |
1da177e4 LT |
657 | |
658 | /* try_to_wake_up() stats */ | |
480b9434 KC |
659 | unsigned int ttwu_count; |
660 | unsigned int ttwu_local; | |
b8efb561 IM |
661 | |
662 | /* BKL stats */ | |
480b9434 | 663 | unsigned int bkl_count; |
1da177e4 LT |
664 | #endif |
665 | }; | |
666 | ||
f34e3b61 | 667 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 668 | |
15afe09b | 669 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 670 | { |
15afe09b | 671 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
672 | } |
673 | ||
0a2966b4 CL |
674 | static inline int cpu_of(struct rq *rq) |
675 | { | |
676 | #ifdef CONFIG_SMP | |
677 | return rq->cpu; | |
678 | #else | |
679 | return 0; | |
680 | #endif | |
681 | } | |
682 | ||
674311d5 NP |
683 | /* |
684 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 685 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
686 | * |
687 | * The domain tree of any CPU may only be accessed from within | |
688 | * preempt-disabled sections. | |
689 | */ | |
48f24c4d IM |
690 | #define for_each_domain(cpu, __sd) \ |
691 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
692 | |
693 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
694 | #define this_rq() (&__get_cpu_var(runqueues)) | |
695 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
696 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 697 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 698 | |
aa9c4c0f | 699 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
700 | { |
701 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
702 | } | |
703 | ||
bf5c91ba IM |
704 | /* |
705 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
706 | */ | |
707 | #ifdef CONFIG_SCHED_DEBUG | |
708 | # define const_debug __read_mostly | |
709 | #else | |
710 | # define const_debug static const | |
711 | #endif | |
712 | ||
017730c1 IM |
713 | /** |
714 | * runqueue_is_locked | |
715 | * | |
716 | * Returns true if the current cpu runqueue is locked. | |
717 | * This interface allows printk to be called with the runqueue lock | |
718 | * held and know whether or not it is OK to wake up the klogd. | |
719 | */ | |
720 | int runqueue_is_locked(void) | |
721 | { | |
722 | int cpu = get_cpu(); | |
723 | struct rq *rq = cpu_rq(cpu); | |
724 | int ret; | |
725 | ||
726 | ret = spin_is_locked(&rq->lock); | |
727 | put_cpu(); | |
728 | return ret; | |
729 | } | |
730 | ||
bf5c91ba IM |
731 | /* |
732 | * Debugging: various feature bits | |
733 | */ | |
f00b45c1 PZ |
734 | |
735 | #define SCHED_FEAT(name, enabled) \ | |
736 | __SCHED_FEAT_##name , | |
737 | ||
bf5c91ba | 738 | enum { |
f00b45c1 | 739 | #include "sched_features.h" |
bf5c91ba IM |
740 | }; |
741 | ||
f00b45c1 PZ |
742 | #undef SCHED_FEAT |
743 | ||
744 | #define SCHED_FEAT(name, enabled) \ | |
745 | (1UL << __SCHED_FEAT_##name) * enabled | | |
746 | ||
bf5c91ba | 747 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
748 | #include "sched_features.h" |
749 | 0; | |
750 | ||
751 | #undef SCHED_FEAT | |
752 | ||
753 | #ifdef CONFIG_SCHED_DEBUG | |
754 | #define SCHED_FEAT(name, enabled) \ | |
755 | #name , | |
756 | ||
983ed7a6 | 757 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
758 | #include "sched_features.h" |
759 | NULL | |
760 | }; | |
761 | ||
762 | #undef SCHED_FEAT | |
763 | ||
34f3a814 | 764 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 765 | { |
f00b45c1 PZ |
766 | int i; |
767 | ||
768 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
769 | if (!(sysctl_sched_features & (1UL << i))) |
770 | seq_puts(m, "NO_"); | |
771 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 772 | } |
34f3a814 | 773 | seq_puts(m, "\n"); |
f00b45c1 | 774 | |
34f3a814 | 775 | return 0; |
f00b45c1 PZ |
776 | } |
777 | ||
778 | static ssize_t | |
779 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
780 | size_t cnt, loff_t *ppos) | |
781 | { | |
782 | char buf[64]; | |
783 | char *cmp = buf; | |
784 | int neg = 0; | |
785 | int i; | |
786 | ||
787 | if (cnt > 63) | |
788 | cnt = 63; | |
789 | ||
790 | if (copy_from_user(&buf, ubuf, cnt)) | |
791 | return -EFAULT; | |
792 | ||
793 | buf[cnt] = 0; | |
794 | ||
c24b7c52 | 795 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
796 | neg = 1; |
797 | cmp += 3; | |
798 | } | |
799 | ||
800 | for (i = 0; sched_feat_names[i]; i++) { | |
801 | int len = strlen(sched_feat_names[i]); | |
802 | ||
803 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
804 | if (neg) | |
805 | sysctl_sched_features &= ~(1UL << i); | |
806 | else | |
807 | sysctl_sched_features |= (1UL << i); | |
808 | break; | |
809 | } | |
810 | } | |
811 | ||
812 | if (!sched_feat_names[i]) | |
813 | return -EINVAL; | |
814 | ||
815 | filp->f_pos += cnt; | |
816 | ||
817 | return cnt; | |
818 | } | |
819 | ||
34f3a814 LZ |
820 | static int sched_feat_open(struct inode *inode, struct file *filp) |
821 | { | |
822 | return single_open(filp, sched_feat_show, NULL); | |
823 | } | |
824 | ||
f00b45c1 | 825 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
826 | .open = sched_feat_open, |
827 | .write = sched_feat_write, | |
828 | .read = seq_read, | |
829 | .llseek = seq_lseek, | |
830 | .release = single_release, | |
f00b45c1 PZ |
831 | }; |
832 | ||
833 | static __init int sched_init_debug(void) | |
834 | { | |
f00b45c1 PZ |
835 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
836 | &sched_feat_fops); | |
837 | ||
838 | return 0; | |
839 | } | |
840 | late_initcall(sched_init_debug); | |
841 | ||
842 | #endif | |
843 | ||
844 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 845 | |
b82d9fdd PZ |
846 | /* |
847 | * Number of tasks to iterate in a single balance run. | |
848 | * Limited because this is done with IRQs disabled. | |
849 | */ | |
850 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
851 | ||
2398f2c6 PZ |
852 | /* |
853 | * ratelimit for updating the group shares. | |
55cd5340 | 854 | * default: 0.25ms |
2398f2c6 | 855 | */ |
55cd5340 | 856 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 857 | |
ffda12a1 PZ |
858 | /* |
859 | * Inject some fuzzyness into changing the per-cpu group shares | |
860 | * this avoids remote rq-locks at the expense of fairness. | |
861 | * default: 4 | |
862 | */ | |
863 | unsigned int sysctl_sched_shares_thresh = 4; | |
864 | ||
fa85ae24 | 865 | /* |
9f0c1e56 | 866 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
867 | * default: 1s |
868 | */ | |
9f0c1e56 | 869 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 870 | |
6892b75e IM |
871 | static __read_mostly int scheduler_running; |
872 | ||
9f0c1e56 PZ |
873 | /* |
874 | * part of the period that we allow rt tasks to run in us. | |
875 | * default: 0.95s | |
876 | */ | |
877 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 878 | |
d0b27fa7 PZ |
879 | static inline u64 global_rt_period(void) |
880 | { | |
881 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
882 | } | |
883 | ||
884 | static inline u64 global_rt_runtime(void) | |
885 | { | |
e26873bb | 886 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
887 | return RUNTIME_INF; |
888 | ||
889 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
890 | } | |
fa85ae24 | 891 | |
1da177e4 | 892 | #ifndef prepare_arch_switch |
4866cde0 NP |
893 | # define prepare_arch_switch(next) do { } while (0) |
894 | #endif | |
895 | #ifndef finish_arch_switch | |
896 | # define finish_arch_switch(prev) do { } while (0) | |
897 | #endif | |
898 | ||
051a1d1a DA |
899 | static inline int task_current(struct rq *rq, struct task_struct *p) |
900 | { | |
901 | return rq->curr == p; | |
902 | } | |
903 | ||
4866cde0 | 904 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 905 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 906 | { |
051a1d1a | 907 | return task_current(rq, p); |
4866cde0 NP |
908 | } |
909 | ||
70b97a7f | 910 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
911 | { |
912 | } | |
913 | ||
70b97a7f | 914 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 915 | { |
da04c035 IM |
916 | #ifdef CONFIG_DEBUG_SPINLOCK |
917 | /* this is a valid case when another task releases the spinlock */ | |
918 | rq->lock.owner = current; | |
919 | #endif | |
8a25d5de IM |
920 | /* |
921 | * If we are tracking spinlock dependencies then we have to | |
922 | * fix up the runqueue lock - which gets 'carried over' from | |
923 | * prev into current: | |
924 | */ | |
925 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
926 | ||
4866cde0 NP |
927 | spin_unlock_irq(&rq->lock); |
928 | } | |
929 | ||
930 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 931 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
932 | { |
933 | #ifdef CONFIG_SMP | |
934 | return p->oncpu; | |
935 | #else | |
051a1d1a | 936 | return task_current(rq, p); |
4866cde0 NP |
937 | #endif |
938 | } | |
939 | ||
70b97a7f | 940 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
941 | { |
942 | #ifdef CONFIG_SMP | |
943 | /* | |
944 | * We can optimise this out completely for !SMP, because the | |
945 | * SMP rebalancing from interrupt is the only thing that cares | |
946 | * here. | |
947 | */ | |
948 | next->oncpu = 1; | |
949 | #endif | |
950 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
951 | spin_unlock_irq(&rq->lock); | |
952 | #else | |
953 | spin_unlock(&rq->lock); | |
954 | #endif | |
955 | } | |
956 | ||
70b97a7f | 957 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
958 | { |
959 | #ifdef CONFIG_SMP | |
960 | /* | |
961 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
962 | * We must ensure this doesn't happen until the switch is completely | |
963 | * finished. | |
964 | */ | |
965 | smp_wmb(); | |
966 | prev->oncpu = 0; | |
967 | #endif | |
968 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
969 | local_irq_enable(); | |
1da177e4 | 970 | #endif |
4866cde0 NP |
971 | } |
972 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 973 | |
b29739f9 IM |
974 | /* |
975 | * __task_rq_lock - lock the runqueue a given task resides on. | |
976 | * Must be called interrupts disabled. | |
977 | */ | |
70b97a7f | 978 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
979 | __acquires(rq->lock) |
980 | { | |
3a5c359a AK |
981 | for (;;) { |
982 | struct rq *rq = task_rq(p); | |
983 | spin_lock(&rq->lock); | |
984 | if (likely(rq == task_rq(p))) | |
985 | return rq; | |
b29739f9 | 986 | spin_unlock(&rq->lock); |
b29739f9 | 987 | } |
b29739f9 IM |
988 | } |
989 | ||
1da177e4 LT |
990 | /* |
991 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 992 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
993 | * explicitly disabling preemption. |
994 | */ | |
70b97a7f | 995 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
996 | __acquires(rq->lock) |
997 | { | |
70b97a7f | 998 | struct rq *rq; |
1da177e4 | 999 | |
3a5c359a AK |
1000 | for (;;) { |
1001 | local_irq_save(*flags); | |
1002 | rq = task_rq(p); | |
1003 | spin_lock(&rq->lock); | |
1004 | if (likely(rq == task_rq(p))) | |
1005 | return rq; | |
1da177e4 | 1006 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1007 | } |
1da177e4 LT |
1008 | } |
1009 | ||
ad474cac ON |
1010 | void task_rq_unlock_wait(struct task_struct *p) |
1011 | { | |
1012 | struct rq *rq = task_rq(p); | |
1013 | ||
1014 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1015 | spin_unlock_wait(&rq->lock); | |
1016 | } | |
1017 | ||
a9957449 | 1018 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1019 | __releases(rq->lock) |
1020 | { | |
1021 | spin_unlock(&rq->lock); | |
1022 | } | |
1023 | ||
70b97a7f | 1024 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1025 | __releases(rq->lock) |
1026 | { | |
1027 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1028 | } | |
1029 | ||
1da177e4 | 1030 | /* |
cc2a73b5 | 1031 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1032 | */ |
a9957449 | 1033 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1034 | __acquires(rq->lock) |
1035 | { | |
70b97a7f | 1036 | struct rq *rq; |
1da177e4 LT |
1037 | |
1038 | local_irq_disable(); | |
1039 | rq = this_rq(); | |
1040 | spin_lock(&rq->lock); | |
1041 | ||
1042 | return rq; | |
1043 | } | |
1044 | ||
8f4d37ec PZ |
1045 | #ifdef CONFIG_SCHED_HRTICK |
1046 | /* | |
1047 | * Use HR-timers to deliver accurate preemption points. | |
1048 | * | |
1049 | * Its all a bit involved since we cannot program an hrt while holding the | |
1050 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1051 | * reschedule event. | |
1052 | * | |
1053 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1054 | * rq->lock. | |
1055 | */ | |
8f4d37ec PZ |
1056 | |
1057 | /* | |
1058 | * Use hrtick when: | |
1059 | * - enabled by features | |
1060 | * - hrtimer is actually high res | |
1061 | */ | |
1062 | static inline int hrtick_enabled(struct rq *rq) | |
1063 | { | |
1064 | if (!sched_feat(HRTICK)) | |
1065 | return 0; | |
ba42059f | 1066 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1067 | return 0; |
8f4d37ec PZ |
1068 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1069 | } | |
1070 | ||
8f4d37ec PZ |
1071 | static void hrtick_clear(struct rq *rq) |
1072 | { | |
1073 | if (hrtimer_active(&rq->hrtick_timer)) | |
1074 | hrtimer_cancel(&rq->hrtick_timer); | |
1075 | } | |
1076 | ||
8f4d37ec PZ |
1077 | /* |
1078 | * High-resolution timer tick. | |
1079 | * Runs from hardirq context with interrupts disabled. | |
1080 | */ | |
1081 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1082 | { | |
1083 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1084 | ||
1085 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1086 | ||
1087 | spin_lock(&rq->lock); | |
3e51f33f | 1088 | update_rq_clock(rq); |
8f4d37ec PZ |
1089 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1090 | spin_unlock(&rq->lock); | |
1091 | ||
1092 | return HRTIMER_NORESTART; | |
1093 | } | |
1094 | ||
95e904c7 | 1095 | #ifdef CONFIG_SMP |
31656519 PZ |
1096 | /* |
1097 | * called from hardirq (IPI) context | |
1098 | */ | |
1099 | static void __hrtick_start(void *arg) | |
b328ca18 | 1100 | { |
31656519 | 1101 | struct rq *rq = arg; |
b328ca18 | 1102 | |
31656519 PZ |
1103 | spin_lock(&rq->lock); |
1104 | hrtimer_restart(&rq->hrtick_timer); | |
1105 | rq->hrtick_csd_pending = 0; | |
1106 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1107 | } |
1108 | ||
31656519 PZ |
1109 | /* |
1110 | * Called to set the hrtick timer state. | |
1111 | * | |
1112 | * called with rq->lock held and irqs disabled | |
1113 | */ | |
1114 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1115 | { |
31656519 PZ |
1116 | struct hrtimer *timer = &rq->hrtick_timer; |
1117 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1118 | |
cc584b21 | 1119 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1120 | |
1121 | if (rq == this_rq()) { | |
1122 | hrtimer_restart(timer); | |
1123 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1124 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1125 | rq->hrtick_csd_pending = 1; |
1126 | } | |
b328ca18 PZ |
1127 | } |
1128 | ||
1129 | static int | |
1130 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1131 | { | |
1132 | int cpu = (int)(long)hcpu; | |
1133 | ||
1134 | switch (action) { | |
1135 | case CPU_UP_CANCELED: | |
1136 | case CPU_UP_CANCELED_FROZEN: | |
1137 | case CPU_DOWN_PREPARE: | |
1138 | case CPU_DOWN_PREPARE_FROZEN: | |
1139 | case CPU_DEAD: | |
1140 | case CPU_DEAD_FROZEN: | |
31656519 | 1141 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1142 | return NOTIFY_OK; |
1143 | } | |
1144 | ||
1145 | return NOTIFY_DONE; | |
1146 | } | |
1147 | ||
fa748203 | 1148 | static __init void init_hrtick(void) |
b328ca18 PZ |
1149 | { |
1150 | hotcpu_notifier(hotplug_hrtick, 0); | |
1151 | } | |
31656519 PZ |
1152 | #else |
1153 | /* | |
1154 | * Called to set the hrtick timer state. | |
1155 | * | |
1156 | * called with rq->lock held and irqs disabled | |
1157 | */ | |
1158 | static void hrtick_start(struct rq *rq, u64 delay) | |
1159 | { | |
7f1e2ca9 | 1160 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1161 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1162 | } |
b328ca18 | 1163 | |
006c75f1 | 1164 | static inline void init_hrtick(void) |
8f4d37ec | 1165 | { |
8f4d37ec | 1166 | } |
31656519 | 1167 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1168 | |
31656519 | 1169 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1170 | { |
31656519 PZ |
1171 | #ifdef CONFIG_SMP |
1172 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1173 | |
31656519 PZ |
1174 | rq->hrtick_csd.flags = 0; |
1175 | rq->hrtick_csd.func = __hrtick_start; | |
1176 | rq->hrtick_csd.info = rq; | |
1177 | #endif | |
8f4d37ec | 1178 | |
31656519 PZ |
1179 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1180 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1181 | } |
006c75f1 | 1182 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1183 | static inline void hrtick_clear(struct rq *rq) |
1184 | { | |
1185 | } | |
1186 | ||
8f4d37ec PZ |
1187 | static inline void init_rq_hrtick(struct rq *rq) |
1188 | { | |
1189 | } | |
1190 | ||
b328ca18 PZ |
1191 | static inline void init_hrtick(void) |
1192 | { | |
1193 | } | |
006c75f1 | 1194 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1195 | |
c24d20db IM |
1196 | /* |
1197 | * resched_task - mark a task 'to be rescheduled now'. | |
1198 | * | |
1199 | * On UP this means the setting of the need_resched flag, on SMP it | |
1200 | * might also involve a cross-CPU call to trigger the scheduler on | |
1201 | * the target CPU. | |
1202 | */ | |
1203 | #ifdef CONFIG_SMP | |
1204 | ||
1205 | #ifndef tsk_is_polling | |
1206 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1207 | #endif | |
1208 | ||
31656519 | 1209 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1210 | { |
1211 | int cpu; | |
1212 | ||
1213 | assert_spin_locked(&task_rq(p)->lock); | |
1214 | ||
5ed0cec0 | 1215 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1216 | return; |
1217 | ||
5ed0cec0 | 1218 | set_tsk_need_resched(p); |
c24d20db IM |
1219 | |
1220 | cpu = task_cpu(p); | |
1221 | if (cpu == smp_processor_id()) | |
1222 | return; | |
1223 | ||
1224 | /* NEED_RESCHED must be visible before we test polling */ | |
1225 | smp_mb(); | |
1226 | if (!tsk_is_polling(p)) | |
1227 | smp_send_reschedule(cpu); | |
1228 | } | |
1229 | ||
1230 | static void resched_cpu(int cpu) | |
1231 | { | |
1232 | struct rq *rq = cpu_rq(cpu); | |
1233 | unsigned long flags; | |
1234 | ||
1235 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1236 | return; | |
1237 | resched_task(cpu_curr(cpu)); | |
1238 | spin_unlock_irqrestore(&rq->lock, flags); | |
1239 | } | |
06d8308c TG |
1240 | |
1241 | #ifdef CONFIG_NO_HZ | |
1242 | /* | |
1243 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1244 | * idle CPU then this timer might expire before the next timer event | |
1245 | * which is scheduled to wake up that CPU. In case of a completely | |
1246 | * idle system the next event might even be infinite time into the | |
1247 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1248 | * leaves the inner idle loop so the newly added timer is taken into | |
1249 | * account when the CPU goes back to idle and evaluates the timer | |
1250 | * wheel for the next timer event. | |
1251 | */ | |
1252 | void wake_up_idle_cpu(int cpu) | |
1253 | { | |
1254 | struct rq *rq = cpu_rq(cpu); | |
1255 | ||
1256 | if (cpu == smp_processor_id()) | |
1257 | return; | |
1258 | ||
1259 | /* | |
1260 | * This is safe, as this function is called with the timer | |
1261 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1262 | * to idle and has not yet set rq->curr to idle then it will | |
1263 | * be serialized on the timer wheel base lock and take the new | |
1264 | * timer into account automatically. | |
1265 | */ | |
1266 | if (rq->curr != rq->idle) | |
1267 | return; | |
1268 | ||
1269 | /* | |
1270 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1271 | * lockless. The worst case is that the other CPU runs the | |
1272 | * idle task through an additional NOOP schedule() | |
1273 | */ | |
5ed0cec0 | 1274 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1275 | |
1276 | /* NEED_RESCHED must be visible before we test polling */ | |
1277 | smp_mb(); | |
1278 | if (!tsk_is_polling(rq->idle)) | |
1279 | smp_send_reschedule(cpu); | |
1280 | } | |
6d6bc0ad | 1281 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1282 | |
6d6bc0ad | 1283 | #else /* !CONFIG_SMP */ |
31656519 | 1284 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1285 | { |
1286 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1287 | set_tsk_need_resched(p); |
c24d20db | 1288 | } |
6d6bc0ad | 1289 | #endif /* CONFIG_SMP */ |
c24d20db | 1290 | |
45bf76df IM |
1291 | #if BITS_PER_LONG == 32 |
1292 | # define WMULT_CONST (~0UL) | |
1293 | #else | |
1294 | # define WMULT_CONST (1UL << 32) | |
1295 | #endif | |
1296 | ||
1297 | #define WMULT_SHIFT 32 | |
1298 | ||
194081eb IM |
1299 | /* |
1300 | * Shift right and round: | |
1301 | */ | |
cf2ab469 | 1302 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1303 | |
a7be37ac PZ |
1304 | /* |
1305 | * delta *= weight / lw | |
1306 | */ | |
cb1c4fc9 | 1307 | static unsigned long |
45bf76df IM |
1308 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1309 | struct load_weight *lw) | |
1310 | { | |
1311 | u64 tmp; | |
1312 | ||
7a232e03 LJ |
1313 | if (!lw->inv_weight) { |
1314 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1315 | lw->inv_weight = 1; | |
1316 | else | |
1317 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1318 | / (lw->weight+1); | |
1319 | } | |
45bf76df IM |
1320 | |
1321 | tmp = (u64)delta_exec * weight; | |
1322 | /* | |
1323 | * Check whether we'd overflow the 64-bit multiplication: | |
1324 | */ | |
194081eb | 1325 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1326 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1327 | WMULT_SHIFT/2); |
1328 | else | |
cf2ab469 | 1329 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1330 | |
ecf691da | 1331 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1332 | } |
1333 | ||
1091985b | 1334 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1335 | { |
1336 | lw->weight += inc; | |
e89996ae | 1337 | lw->inv_weight = 0; |
45bf76df IM |
1338 | } |
1339 | ||
1091985b | 1340 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1341 | { |
1342 | lw->weight -= dec; | |
e89996ae | 1343 | lw->inv_weight = 0; |
45bf76df IM |
1344 | } |
1345 | ||
2dd73a4f PW |
1346 | /* |
1347 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1348 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1349 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1350 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1351 | * scaled version of the new time slice allocation that they receive on time |
1352 | * slice expiry etc. | |
1353 | */ | |
1354 | ||
cce7ade8 PZ |
1355 | #define WEIGHT_IDLEPRIO 3 |
1356 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1357 | |
1358 | /* | |
1359 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1360 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1361 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1362 | * that remained on nice 0. | |
1363 | * | |
1364 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1365 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1366 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1367 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1368 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1369 | */ |
1370 | static const int prio_to_weight[40] = { | |
254753dc IM |
1371 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1372 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1373 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1374 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1375 | /* 0 */ 1024, 820, 655, 526, 423, | |
1376 | /* 5 */ 335, 272, 215, 172, 137, | |
1377 | /* 10 */ 110, 87, 70, 56, 45, | |
1378 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1379 | }; |
1380 | ||
5714d2de IM |
1381 | /* |
1382 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1383 | * | |
1384 | * In cases where the weight does not change often, we can use the | |
1385 | * precalculated inverse to speed up arithmetics by turning divisions | |
1386 | * into multiplications: | |
1387 | */ | |
dd41f596 | 1388 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1389 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1390 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1391 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1392 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1393 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1394 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1395 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1396 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1397 | }; |
2dd73a4f | 1398 | |
dd41f596 IM |
1399 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1400 | ||
1401 | /* | |
1402 | * runqueue iterator, to support SMP load-balancing between different | |
1403 | * scheduling classes, without having to expose their internal data | |
1404 | * structures to the load-balancing proper: | |
1405 | */ | |
1406 | struct rq_iterator { | |
1407 | void *arg; | |
1408 | struct task_struct *(*start)(void *); | |
1409 | struct task_struct *(*next)(void *); | |
1410 | }; | |
1411 | ||
e1d1484f PW |
1412 | #ifdef CONFIG_SMP |
1413 | static unsigned long | |
1414 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1415 | unsigned long max_load_move, struct sched_domain *sd, | |
1416 | enum cpu_idle_type idle, int *all_pinned, | |
1417 | int *this_best_prio, struct rq_iterator *iterator); | |
1418 | ||
1419 | static int | |
1420 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1421 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1422 | struct rq_iterator *iterator); | |
e1d1484f | 1423 | #endif |
dd41f596 | 1424 | |
ef12fefa BR |
1425 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1426 | enum cpuacct_stat_index { | |
1427 | CPUACCT_STAT_USER, /* ... user mode */ | |
1428 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1429 | ||
1430 | CPUACCT_STAT_NSTATS, | |
1431 | }; | |
1432 | ||
d842de87 SV |
1433 | #ifdef CONFIG_CGROUP_CPUACCT |
1434 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1435 | static void cpuacct_update_stats(struct task_struct *tsk, |
1436 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1437 | #else |
1438 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1439 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1440 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1441 | #endif |
1442 | ||
18d95a28 PZ |
1443 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1444 | { | |
1445 | update_load_add(&rq->load, load); | |
1446 | } | |
1447 | ||
1448 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1449 | { | |
1450 | update_load_sub(&rq->load, load); | |
1451 | } | |
1452 | ||
7940ca36 | 1453 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1454 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1455 | |
1456 | /* | |
1457 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1458 | * leaving it for the final time. | |
1459 | */ | |
eb755805 | 1460 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1461 | { |
1462 | struct task_group *parent, *child; | |
eb755805 | 1463 | int ret; |
c09595f6 PZ |
1464 | |
1465 | rcu_read_lock(); | |
1466 | parent = &root_task_group; | |
1467 | down: | |
eb755805 PZ |
1468 | ret = (*down)(parent, data); |
1469 | if (ret) | |
1470 | goto out_unlock; | |
c09595f6 PZ |
1471 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1472 | parent = child; | |
1473 | goto down; | |
1474 | ||
1475 | up: | |
1476 | continue; | |
1477 | } | |
eb755805 PZ |
1478 | ret = (*up)(parent, data); |
1479 | if (ret) | |
1480 | goto out_unlock; | |
c09595f6 PZ |
1481 | |
1482 | child = parent; | |
1483 | parent = parent->parent; | |
1484 | if (parent) | |
1485 | goto up; | |
eb755805 | 1486 | out_unlock: |
c09595f6 | 1487 | rcu_read_unlock(); |
eb755805 PZ |
1488 | |
1489 | return ret; | |
c09595f6 PZ |
1490 | } |
1491 | ||
eb755805 PZ |
1492 | static int tg_nop(struct task_group *tg, void *data) |
1493 | { | |
1494 | return 0; | |
c09595f6 | 1495 | } |
eb755805 PZ |
1496 | #endif |
1497 | ||
1498 | #ifdef CONFIG_SMP | |
1499 | static unsigned long source_load(int cpu, int type); | |
1500 | static unsigned long target_load(int cpu, int type); | |
1501 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1502 | ||
1503 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1504 | { | |
1505 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1506 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1507 | |
4cd42620 SR |
1508 | if (nr_running) |
1509 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1510 | else |
1511 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1512 | |
1513 | return rq->avg_load_per_task; | |
1514 | } | |
1515 | ||
1516 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1517 | |
c09595f6 PZ |
1518 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1519 | ||
1520 | /* | |
1521 | * Calculate and set the cpu's group shares. | |
1522 | */ | |
1523 | static void | |
ffda12a1 | 1524 | update_group_shares_cpu(struct task_group *tg, int cpu, |
a8af7246 PZ |
1525 | unsigned long sd_shares, unsigned long sd_rq_weight, |
1526 | unsigned long sd_eff_weight) | |
18d95a28 | 1527 | { |
c09595f6 | 1528 | unsigned long rq_weight; |
a5004278 PZ |
1529 | unsigned long shares; |
1530 | int boost = 0; | |
c09595f6 | 1531 | |
c8cba857 | 1532 | if (!tg->se[cpu]) |
c09595f6 PZ |
1533 | return; |
1534 | ||
ec4e0e2f | 1535 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
a5004278 PZ |
1536 | if (!rq_weight) { |
1537 | boost = 1; | |
1538 | rq_weight = NICE_0_LOAD; | |
a8af7246 PZ |
1539 | if (sd_rq_weight == sd_eff_weight) |
1540 | sd_eff_weight += NICE_0_LOAD; | |
1541 | sd_rq_weight = sd_eff_weight; | |
a5004278 | 1542 | } |
c8cba857 | 1543 | |
c09595f6 | 1544 | /* |
a8af7246 PZ |
1545 | * \Sum_j shares_j * rq_weight_i |
1546 | * shares_i = ----------------------------- | |
1547 | * \Sum_j rq_weight_j | |
c09595f6 | 1548 | */ |
ec4e0e2f | 1549 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1550 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1551 | |
ffda12a1 PZ |
1552 | if (abs(shares - tg->se[cpu]->load.weight) > |
1553 | sysctl_sched_shares_thresh) { | |
1554 | struct rq *rq = cpu_rq(cpu); | |
1555 | unsigned long flags; | |
c09595f6 | 1556 | |
ffda12a1 | 1557 | spin_lock_irqsave(&rq->lock, flags); |
a5004278 | 1558 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 PZ |
1559 | __set_se_shares(tg->se[cpu], shares); |
1560 | spin_unlock_irqrestore(&rq->lock, flags); | |
1561 | } | |
18d95a28 | 1562 | } |
c09595f6 PZ |
1563 | |
1564 | /* | |
c8cba857 PZ |
1565 | * Re-compute the task group their per cpu shares over the given domain. |
1566 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1567 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1568 | */ |
eb755805 | 1569 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1570 | { |
a5004278 | 1571 | unsigned long weight, rq_weight = 0, eff_weight = 0; |
c8cba857 | 1572 | unsigned long shares = 0; |
eb755805 | 1573 | struct sched_domain *sd = data; |
c8cba857 | 1574 | int i; |
c09595f6 | 1575 | |
758b2cdc | 1576 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1577 | /* |
1578 | * If there are currently no tasks on the cpu pretend there | |
1579 | * is one of average load so that when a new task gets to | |
1580 | * run here it will not get delayed by group starvation. | |
1581 | */ | |
1582 | weight = tg->cfs_rq[i]->load.weight; | |
a5004278 PZ |
1583 | tg->cfs_rq[i]->rq_weight = weight; |
1584 | rq_weight += weight; | |
1585 | ||
ec4e0e2f KC |
1586 | if (!weight) |
1587 | weight = NICE_0_LOAD; | |
1588 | ||
a5004278 | 1589 | eff_weight += weight; |
c8cba857 | 1590 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1591 | } |
c09595f6 | 1592 | |
c8cba857 PZ |
1593 | if ((!shares && rq_weight) || shares > tg->shares) |
1594 | shares = tg->shares; | |
1595 | ||
1596 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1597 | shares = tg->shares; | |
c09595f6 | 1598 | |
a8af7246 PZ |
1599 | for_each_cpu(i, sched_domain_span(sd)) |
1600 | update_group_shares_cpu(tg, i, shares, rq_weight, eff_weight); | |
eb755805 PZ |
1601 | |
1602 | return 0; | |
c09595f6 PZ |
1603 | } |
1604 | ||
1605 | /* | |
c8cba857 PZ |
1606 | * Compute the cpu's hierarchical load factor for each task group. |
1607 | * This needs to be done in a top-down fashion because the load of a child | |
1608 | * group is a fraction of its parents load. | |
c09595f6 | 1609 | */ |
eb755805 | 1610 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1611 | { |
c8cba857 | 1612 | unsigned long load; |
eb755805 | 1613 | long cpu = (long)data; |
c09595f6 | 1614 | |
c8cba857 PZ |
1615 | if (!tg->parent) { |
1616 | load = cpu_rq(cpu)->load.weight; | |
1617 | } else { | |
1618 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1619 | load *= tg->cfs_rq[cpu]->shares; | |
1620 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1621 | } | |
c09595f6 | 1622 | |
c8cba857 | 1623 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1624 | |
eb755805 | 1625 | return 0; |
c09595f6 PZ |
1626 | } |
1627 | ||
c8cba857 | 1628 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1629 | { |
e7097159 PZ |
1630 | s64 elapsed; |
1631 | u64 now; | |
1632 | ||
1633 | if (root_task_group_empty()) | |
1634 | return; | |
1635 | ||
1636 | now = cpu_clock(raw_smp_processor_id()); | |
1637 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1638 | |
1639 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1640 | sd->last_update = now; | |
eb755805 | 1641 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1642 | } |
4d8d595d PZ |
1643 | } |
1644 | ||
3e5459b4 PZ |
1645 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1646 | { | |
e7097159 PZ |
1647 | if (root_task_group_empty()) |
1648 | return; | |
1649 | ||
3e5459b4 PZ |
1650 | spin_unlock(&rq->lock); |
1651 | update_shares(sd); | |
1652 | spin_lock(&rq->lock); | |
1653 | } | |
1654 | ||
eb755805 | 1655 | static void update_h_load(long cpu) |
c09595f6 | 1656 | { |
e7097159 PZ |
1657 | if (root_task_group_empty()) |
1658 | return; | |
1659 | ||
eb755805 | 1660 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1661 | } |
1662 | ||
c09595f6 PZ |
1663 | #else |
1664 | ||
c8cba857 | 1665 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1666 | { |
1667 | } | |
1668 | ||
3e5459b4 PZ |
1669 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1670 | { | |
1671 | } | |
1672 | ||
18d95a28 PZ |
1673 | #endif |
1674 | ||
8f45e2b5 GH |
1675 | #ifdef CONFIG_PREEMPT |
1676 | ||
70574a99 | 1677 | /* |
8f45e2b5 GH |
1678 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1679 | * way at the expense of forcing extra atomic operations in all | |
1680 | * invocations. This assures that the double_lock is acquired using the | |
1681 | * same underlying policy as the spinlock_t on this architecture, which | |
1682 | * reduces latency compared to the unfair variant below. However, it | |
1683 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1684 | */ |
8f45e2b5 GH |
1685 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1686 | __releases(this_rq->lock) | |
1687 | __acquires(busiest->lock) | |
1688 | __acquires(this_rq->lock) | |
1689 | { | |
1690 | spin_unlock(&this_rq->lock); | |
1691 | double_rq_lock(this_rq, busiest); | |
1692 | ||
1693 | return 1; | |
1694 | } | |
1695 | ||
1696 | #else | |
1697 | /* | |
1698 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1699 | * latency by eliminating extra atomic operations when the locks are | |
1700 | * already in proper order on entry. This favors lower cpu-ids and will | |
1701 | * grant the double lock to lower cpus over higher ids under contention, | |
1702 | * regardless of entry order into the function. | |
1703 | */ | |
1704 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1705 | __releases(this_rq->lock) |
1706 | __acquires(busiest->lock) | |
1707 | __acquires(this_rq->lock) | |
1708 | { | |
1709 | int ret = 0; | |
1710 | ||
70574a99 AD |
1711 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1712 | if (busiest < this_rq) { | |
1713 | spin_unlock(&this_rq->lock); | |
1714 | spin_lock(&busiest->lock); | |
1715 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1716 | ret = 1; | |
1717 | } else | |
1718 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1719 | } | |
1720 | return ret; | |
1721 | } | |
1722 | ||
8f45e2b5 GH |
1723 | #endif /* CONFIG_PREEMPT */ |
1724 | ||
1725 | /* | |
1726 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1727 | */ | |
1728 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1729 | { | |
1730 | if (unlikely(!irqs_disabled())) { | |
1731 | /* printk() doesn't work good under rq->lock */ | |
1732 | spin_unlock(&this_rq->lock); | |
1733 | BUG_ON(1); | |
1734 | } | |
1735 | ||
1736 | return _double_lock_balance(this_rq, busiest); | |
1737 | } | |
1738 | ||
70574a99 AD |
1739 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1740 | __releases(busiest->lock) | |
1741 | { | |
1742 | spin_unlock(&busiest->lock); | |
1743 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1744 | } | |
18d95a28 PZ |
1745 | #endif |
1746 | ||
30432094 | 1747 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1748 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1749 | { | |
30432094 | 1750 | #ifdef CONFIG_SMP |
34e83e85 IM |
1751 | cfs_rq->shares = shares; |
1752 | #endif | |
1753 | } | |
30432094 | 1754 | #endif |
e7693a36 | 1755 | |
dce48a84 TG |
1756 | static void calc_load_account_active(struct rq *this_rq); |
1757 | ||
dd41f596 | 1758 | #include "sched_stats.h" |
dd41f596 | 1759 | #include "sched_idletask.c" |
5522d5d5 IM |
1760 | #include "sched_fair.c" |
1761 | #include "sched_rt.c" | |
dd41f596 IM |
1762 | #ifdef CONFIG_SCHED_DEBUG |
1763 | # include "sched_debug.c" | |
1764 | #endif | |
1765 | ||
1766 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1767 | #define for_each_class(class) \ |
1768 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1769 | |
c09595f6 | 1770 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1771 | { |
1772 | rq->nr_running++; | |
9c217245 IM |
1773 | } |
1774 | ||
c09595f6 | 1775 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1776 | { |
1777 | rq->nr_running--; | |
9c217245 IM |
1778 | } |
1779 | ||
45bf76df IM |
1780 | static void set_load_weight(struct task_struct *p) |
1781 | { | |
1782 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1783 | p->se.load.weight = prio_to_weight[0] * 2; |
1784 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1785 | return; | |
1786 | } | |
45bf76df | 1787 | |
dd41f596 IM |
1788 | /* |
1789 | * SCHED_IDLE tasks get minimal weight: | |
1790 | */ | |
1791 | if (p->policy == SCHED_IDLE) { | |
1792 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1793 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1794 | return; | |
1795 | } | |
71f8bd46 | 1796 | |
dd41f596 IM |
1797 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1798 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1799 | } |
1800 | ||
2087a1ad GH |
1801 | static void update_avg(u64 *avg, u64 sample) |
1802 | { | |
1803 | s64 diff = sample - *avg; | |
1804 | *avg += diff >> 3; | |
1805 | } | |
1806 | ||
8159f87e | 1807 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1808 | { |
831451ac PZ |
1809 | if (wakeup) |
1810 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1811 | ||
dd41f596 | 1812 | sched_info_queued(p); |
fd390f6a | 1813 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1814 | p->se.on_rq = 1; |
71f8bd46 IM |
1815 | } |
1816 | ||
69be72c1 | 1817 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1818 | { |
831451ac PZ |
1819 | if (sleep) { |
1820 | if (p->se.last_wakeup) { | |
1821 | update_avg(&p->se.avg_overlap, | |
1822 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1823 | p->se.last_wakeup = 0; | |
1824 | } else { | |
1825 | update_avg(&p->se.avg_wakeup, | |
1826 | sysctl_sched_wakeup_granularity); | |
1827 | } | |
2087a1ad GH |
1828 | } |
1829 | ||
46ac22ba | 1830 | sched_info_dequeued(p); |
f02231e5 | 1831 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1832 | p->se.on_rq = 0; |
71f8bd46 IM |
1833 | } |
1834 | ||
14531189 | 1835 | /* |
dd41f596 | 1836 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1837 | */ |
14531189 IM |
1838 | static inline int __normal_prio(struct task_struct *p) |
1839 | { | |
dd41f596 | 1840 | return p->static_prio; |
14531189 IM |
1841 | } |
1842 | ||
b29739f9 IM |
1843 | /* |
1844 | * Calculate the expected normal priority: i.e. priority | |
1845 | * without taking RT-inheritance into account. Might be | |
1846 | * boosted by interactivity modifiers. Changes upon fork, | |
1847 | * setprio syscalls, and whenever the interactivity | |
1848 | * estimator recalculates. | |
1849 | */ | |
36c8b586 | 1850 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1851 | { |
1852 | int prio; | |
1853 | ||
e05606d3 | 1854 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1855 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1856 | else | |
1857 | prio = __normal_prio(p); | |
1858 | return prio; | |
1859 | } | |
1860 | ||
1861 | /* | |
1862 | * Calculate the current priority, i.e. the priority | |
1863 | * taken into account by the scheduler. This value might | |
1864 | * be boosted by RT tasks, or might be boosted by | |
1865 | * interactivity modifiers. Will be RT if the task got | |
1866 | * RT-boosted. If not then it returns p->normal_prio. | |
1867 | */ | |
36c8b586 | 1868 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1869 | { |
1870 | p->normal_prio = normal_prio(p); | |
1871 | /* | |
1872 | * If we are RT tasks or we were boosted to RT priority, | |
1873 | * keep the priority unchanged. Otherwise, update priority | |
1874 | * to the normal priority: | |
1875 | */ | |
1876 | if (!rt_prio(p->prio)) | |
1877 | return p->normal_prio; | |
1878 | return p->prio; | |
1879 | } | |
1880 | ||
1da177e4 | 1881 | /* |
dd41f596 | 1882 | * activate_task - move a task to the runqueue. |
1da177e4 | 1883 | */ |
dd41f596 | 1884 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1885 | { |
d9514f6c | 1886 | if (task_contributes_to_load(p)) |
dd41f596 | 1887 | rq->nr_uninterruptible--; |
1da177e4 | 1888 | |
8159f87e | 1889 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1890 | inc_nr_running(rq); |
1da177e4 LT |
1891 | } |
1892 | ||
1da177e4 LT |
1893 | /* |
1894 | * deactivate_task - remove a task from the runqueue. | |
1895 | */ | |
2e1cb74a | 1896 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1897 | { |
d9514f6c | 1898 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1899 | rq->nr_uninterruptible++; |
1900 | ||
69be72c1 | 1901 | dequeue_task(rq, p, sleep); |
c09595f6 | 1902 | dec_nr_running(rq); |
1da177e4 LT |
1903 | } |
1904 | ||
1da177e4 LT |
1905 | /** |
1906 | * task_curr - is this task currently executing on a CPU? | |
1907 | * @p: the task in question. | |
1908 | */ | |
36c8b586 | 1909 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1910 | { |
1911 | return cpu_curr(task_cpu(p)) == p; | |
1912 | } | |
1913 | ||
dd41f596 IM |
1914 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1915 | { | |
6f505b16 | 1916 | set_task_rq(p, cpu); |
dd41f596 | 1917 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1918 | /* |
1919 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1920 | * successfuly executed on another CPU. We must ensure that updates of | |
1921 | * per-task data have been completed by this moment. | |
1922 | */ | |
1923 | smp_wmb(); | |
dd41f596 | 1924 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1925 | #endif |
2dd73a4f PW |
1926 | } |
1927 | ||
cb469845 SR |
1928 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1929 | const struct sched_class *prev_class, | |
1930 | int oldprio, int running) | |
1931 | { | |
1932 | if (prev_class != p->sched_class) { | |
1933 | if (prev_class->switched_from) | |
1934 | prev_class->switched_from(rq, p, running); | |
1935 | p->sched_class->switched_to(rq, p, running); | |
1936 | } else | |
1937 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1938 | } | |
1939 | ||
1da177e4 | 1940 | #ifdef CONFIG_SMP |
c65cc870 | 1941 | |
e958b360 TG |
1942 | /* Used instead of source_load when we know the type == 0 */ |
1943 | static unsigned long weighted_cpuload(const int cpu) | |
1944 | { | |
1945 | return cpu_rq(cpu)->load.weight; | |
1946 | } | |
1947 | ||
cc367732 IM |
1948 | /* |
1949 | * Is this task likely cache-hot: | |
1950 | */ | |
e7693a36 | 1951 | static int |
cc367732 IM |
1952 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1953 | { | |
1954 | s64 delta; | |
1955 | ||
f540a608 IM |
1956 | /* |
1957 | * Buddy candidates are cache hot: | |
1958 | */ | |
4793241b PZ |
1959 | if (sched_feat(CACHE_HOT_BUDDY) && |
1960 | (&p->se == cfs_rq_of(&p->se)->next || | |
1961 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1962 | return 1; |
1963 | ||
cc367732 IM |
1964 | if (p->sched_class != &fair_sched_class) |
1965 | return 0; | |
1966 | ||
6bc1665b IM |
1967 | if (sysctl_sched_migration_cost == -1) |
1968 | return 1; | |
1969 | if (sysctl_sched_migration_cost == 0) | |
1970 | return 0; | |
1971 | ||
cc367732 IM |
1972 | delta = now - p->se.exec_start; |
1973 | ||
1974 | return delta < (s64)sysctl_sched_migration_cost; | |
1975 | } | |
1976 | ||
1977 | ||
dd41f596 | 1978 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1979 | { |
dd41f596 IM |
1980 | int old_cpu = task_cpu(p); |
1981 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1982 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1983 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1984 | u64 clock_offset; |
dd41f596 IM |
1985 | |
1986 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1987 | |
de1d7286 | 1988 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1989 | |
6cfb0d5d IM |
1990 | #ifdef CONFIG_SCHEDSTATS |
1991 | if (p->se.wait_start) | |
1992 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1993 | if (p->se.sleep_start) |
1994 | p->se.sleep_start -= clock_offset; | |
1995 | if (p->se.block_start) | |
1996 | p->se.block_start -= clock_offset; | |
6c594c21 | 1997 | #endif |
cc367732 | 1998 | if (old_cpu != new_cpu) { |
6c594c21 | 1999 | p->se.nr_migrations++; |
23a185ca | 2000 | new_rq->nr_migrations_in++; |
6c594c21 | 2001 | #ifdef CONFIG_SCHEDSTATS |
cc367732 IM |
2002 | if (task_hot(p, old_rq->clock, NULL)) |
2003 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 2004 | #endif |
e5289d4a PZ |
2005 | perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
2006 | 1, 1, NULL, 0); | |
6c594c21 | 2007 | } |
2830cf8c SV |
2008 | p->se.vruntime -= old_cfsrq->min_vruntime - |
2009 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
2010 | |
2011 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2012 | } |
2013 | ||
70b97a7f | 2014 | struct migration_req { |
1da177e4 | 2015 | struct list_head list; |
1da177e4 | 2016 | |
36c8b586 | 2017 | struct task_struct *task; |
1da177e4 LT |
2018 | int dest_cpu; |
2019 | ||
1da177e4 | 2020 | struct completion done; |
70b97a7f | 2021 | }; |
1da177e4 LT |
2022 | |
2023 | /* | |
2024 | * The task's runqueue lock must be held. | |
2025 | * Returns true if you have to wait for migration thread. | |
2026 | */ | |
36c8b586 | 2027 | static int |
70b97a7f | 2028 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2029 | { |
70b97a7f | 2030 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2031 | |
2032 | /* | |
2033 | * If the task is not on a runqueue (and not running), then | |
2034 | * it is sufficient to simply update the task's cpu field. | |
2035 | */ | |
dd41f596 | 2036 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2037 | set_task_cpu(p, dest_cpu); |
2038 | return 0; | |
2039 | } | |
2040 | ||
2041 | init_completion(&req->done); | |
1da177e4 LT |
2042 | req->task = p; |
2043 | req->dest_cpu = dest_cpu; | |
2044 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2045 | |
1da177e4 LT |
2046 | return 1; |
2047 | } | |
2048 | ||
a26b89f0 MM |
2049 | /* |
2050 | * wait_task_context_switch - wait for a thread to complete at least one | |
2051 | * context switch. | |
2052 | * | |
2053 | * @p must not be current. | |
2054 | */ | |
2055 | void wait_task_context_switch(struct task_struct *p) | |
2056 | { | |
2057 | unsigned long nvcsw, nivcsw, flags; | |
2058 | int running; | |
2059 | struct rq *rq; | |
2060 | ||
2061 | nvcsw = p->nvcsw; | |
2062 | nivcsw = p->nivcsw; | |
2063 | for (;;) { | |
2064 | /* | |
2065 | * The runqueue is assigned before the actual context | |
2066 | * switch. We need to take the runqueue lock. | |
2067 | * | |
2068 | * We could check initially without the lock but it is | |
2069 | * very likely that we need to take the lock in every | |
2070 | * iteration. | |
2071 | */ | |
2072 | rq = task_rq_lock(p, &flags); | |
2073 | running = task_running(rq, p); | |
2074 | task_rq_unlock(rq, &flags); | |
2075 | ||
2076 | if (likely(!running)) | |
2077 | break; | |
2078 | /* | |
2079 | * The switch count is incremented before the actual | |
2080 | * context switch. We thus wait for two switches to be | |
2081 | * sure at least one completed. | |
2082 | */ | |
2083 | if ((p->nvcsw - nvcsw) > 1) | |
2084 | break; | |
2085 | if ((p->nivcsw - nivcsw) > 1) | |
2086 | break; | |
2087 | ||
2088 | cpu_relax(); | |
2089 | } | |
2090 | } | |
2091 | ||
1da177e4 LT |
2092 | /* |
2093 | * wait_task_inactive - wait for a thread to unschedule. | |
2094 | * | |
85ba2d86 RM |
2095 | * If @match_state is nonzero, it's the @p->state value just checked and |
2096 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2097 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2098 | * we return a positive number (its total switch count). If a second call | |
2099 | * a short while later returns the same number, the caller can be sure that | |
2100 | * @p has remained unscheduled the whole time. | |
2101 | * | |
1da177e4 LT |
2102 | * The caller must ensure that the task *will* unschedule sometime soon, |
2103 | * else this function might spin for a *long* time. This function can't | |
2104 | * be called with interrupts off, or it may introduce deadlock with | |
2105 | * smp_call_function() if an IPI is sent by the same process we are | |
2106 | * waiting to become inactive. | |
2107 | */ | |
85ba2d86 | 2108 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2109 | { |
2110 | unsigned long flags; | |
dd41f596 | 2111 | int running, on_rq; |
85ba2d86 | 2112 | unsigned long ncsw; |
70b97a7f | 2113 | struct rq *rq; |
1da177e4 | 2114 | |
3a5c359a AK |
2115 | for (;;) { |
2116 | /* | |
2117 | * We do the initial early heuristics without holding | |
2118 | * any task-queue locks at all. We'll only try to get | |
2119 | * the runqueue lock when things look like they will | |
2120 | * work out! | |
2121 | */ | |
2122 | rq = task_rq(p); | |
fa490cfd | 2123 | |
3a5c359a AK |
2124 | /* |
2125 | * If the task is actively running on another CPU | |
2126 | * still, just relax and busy-wait without holding | |
2127 | * any locks. | |
2128 | * | |
2129 | * NOTE! Since we don't hold any locks, it's not | |
2130 | * even sure that "rq" stays as the right runqueue! | |
2131 | * But we don't care, since "task_running()" will | |
2132 | * return false if the runqueue has changed and p | |
2133 | * is actually now running somewhere else! | |
2134 | */ | |
85ba2d86 RM |
2135 | while (task_running(rq, p)) { |
2136 | if (match_state && unlikely(p->state != match_state)) | |
2137 | return 0; | |
3a5c359a | 2138 | cpu_relax(); |
85ba2d86 | 2139 | } |
fa490cfd | 2140 | |
3a5c359a AK |
2141 | /* |
2142 | * Ok, time to look more closely! We need the rq | |
2143 | * lock now, to be *sure*. If we're wrong, we'll | |
2144 | * just go back and repeat. | |
2145 | */ | |
2146 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2147 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2148 | running = task_running(rq, p); |
2149 | on_rq = p->se.on_rq; | |
85ba2d86 | 2150 | ncsw = 0; |
f31e11d8 | 2151 | if (!match_state || p->state == match_state) |
93dcf55f | 2152 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2153 | task_rq_unlock(rq, &flags); |
fa490cfd | 2154 | |
85ba2d86 RM |
2155 | /* |
2156 | * If it changed from the expected state, bail out now. | |
2157 | */ | |
2158 | if (unlikely(!ncsw)) | |
2159 | break; | |
2160 | ||
3a5c359a AK |
2161 | /* |
2162 | * Was it really running after all now that we | |
2163 | * checked with the proper locks actually held? | |
2164 | * | |
2165 | * Oops. Go back and try again.. | |
2166 | */ | |
2167 | if (unlikely(running)) { | |
2168 | cpu_relax(); | |
2169 | continue; | |
2170 | } | |
fa490cfd | 2171 | |
3a5c359a AK |
2172 | /* |
2173 | * It's not enough that it's not actively running, | |
2174 | * it must be off the runqueue _entirely_, and not | |
2175 | * preempted! | |
2176 | * | |
80dd99b3 | 2177 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2178 | * running right now), it's preempted, and we should |
2179 | * yield - it could be a while. | |
2180 | */ | |
2181 | if (unlikely(on_rq)) { | |
2182 | schedule_timeout_uninterruptible(1); | |
2183 | continue; | |
2184 | } | |
fa490cfd | 2185 | |
3a5c359a AK |
2186 | /* |
2187 | * Ahh, all good. It wasn't running, and it wasn't | |
2188 | * runnable, which means that it will never become | |
2189 | * running in the future either. We're all done! | |
2190 | */ | |
2191 | break; | |
2192 | } | |
85ba2d86 RM |
2193 | |
2194 | return ncsw; | |
1da177e4 LT |
2195 | } |
2196 | ||
2197 | /*** | |
2198 | * kick_process - kick a running thread to enter/exit the kernel | |
2199 | * @p: the to-be-kicked thread | |
2200 | * | |
2201 | * Cause a process which is running on another CPU to enter | |
2202 | * kernel-mode, without any delay. (to get signals handled.) | |
2203 | * | |
2204 | * NOTE: this function doesnt have to take the runqueue lock, | |
2205 | * because all it wants to ensure is that the remote task enters | |
2206 | * the kernel. If the IPI races and the task has been migrated | |
2207 | * to another CPU then no harm is done and the purpose has been | |
2208 | * achieved as well. | |
2209 | */ | |
36c8b586 | 2210 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2211 | { |
2212 | int cpu; | |
2213 | ||
2214 | preempt_disable(); | |
2215 | cpu = task_cpu(p); | |
2216 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2217 | smp_send_reschedule(cpu); | |
2218 | preempt_enable(); | |
2219 | } | |
b43e3521 | 2220 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 LT |
2221 | |
2222 | /* | |
2dd73a4f PW |
2223 | * Return a low guess at the load of a migration-source cpu weighted |
2224 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2225 | * |
2226 | * We want to under-estimate the load of migration sources, to | |
2227 | * balance conservatively. | |
2228 | */ | |
a9957449 | 2229 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2230 | { |
70b97a7f | 2231 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2232 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2233 | |
93b75217 | 2234 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2235 | return total; |
b910472d | 2236 | |
dd41f596 | 2237 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2238 | } |
2239 | ||
2240 | /* | |
2dd73a4f PW |
2241 | * Return a high guess at the load of a migration-target cpu weighted |
2242 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2243 | */ |
a9957449 | 2244 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2245 | { |
70b97a7f | 2246 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2247 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2248 | |
93b75217 | 2249 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2250 | return total; |
3b0bd9bc | 2251 | |
dd41f596 | 2252 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2253 | } |
2254 | ||
147cbb4b NP |
2255 | /* |
2256 | * find_idlest_group finds and returns the least busy CPU group within the | |
2257 | * domain. | |
2258 | */ | |
2259 | static struct sched_group * | |
2260 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2261 | { | |
2262 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2263 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2264 | int load_idx = sd->forkexec_idx; | |
2265 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2266 | ||
2267 | do { | |
2268 | unsigned long load, avg_load; | |
2269 | int local_group; | |
2270 | int i; | |
2271 | ||
da5a5522 | 2272 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2273 | if (!cpumask_intersects(sched_group_cpus(group), |
2274 | &p->cpus_allowed)) | |
3a5c359a | 2275 | continue; |
da5a5522 | 2276 | |
758b2cdc RR |
2277 | local_group = cpumask_test_cpu(this_cpu, |
2278 | sched_group_cpus(group)); | |
147cbb4b NP |
2279 | |
2280 | /* Tally up the load of all CPUs in the group */ | |
2281 | avg_load = 0; | |
2282 | ||
758b2cdc | 2283 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2284 | /* Bias balancing toward cpus of our domain */ |
2285 | if (local_group) | |
2286 | load = source_load(i, load_idx); | |
2287 | else | |
2288 | load = target_load(i, load_idx); | |
2289 | ||
2290 | avg_load += load; | |
2291 | } | |
2292 | ||
2293 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2294 | avg_load = sg_div_cpu_power(group, |
2295 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2296 | |
2297 | if (local_group) { | |
2298 | this_load = avg_load; | |
2299 | this = group; | |
2300 | } else if (avg_load < min_load) { | |
2301 | min_load = avg_load; | |
2302 | idlest = group; | |
2303 | } | |
3a5c359a | 2304 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2305 | |
2306 | if (!idlest || 100*this_load < imbalance*min_load) | |
2307 | return NULL; | |
2308 | return idlest; | |
2309 | } | |
2310 | ||
2311 | /* | |
0feaece9 | 2312 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2313 | */ |
95cdf3b7 | 2314 | static int |
758b2cdc | 2315 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2316 | { |
2317 | unsigned long load, min_load = ULONG_MAX; | |
2318 | int idlest = -1; | |
2319 | int i; | |
2320 | ||
da5a5522 | 2321 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2322 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2323 | load = weighted_cpuload(i); |
147cbb4b NP |
2324 | |
2325 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2326 | min_load = load; | |
2327 | idlest = i; | |
2328 | } | |
2329 | } | |
2330 | ||
2331 | return idlest; | |
2332 | } | |
2333 | ||
476d139c NP |
2334 | /* |
2335 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2336 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2337 | * SD_BALANCE_EXEC. | |
2338 | * | |
2339 | * Balance, ie. select the least loaded group. | |
2340 | * | |
2341 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2342 | * | |
2343 | * preempt must be disabled. | |
2344 | */ | |
2345 | static int sched_balance_self(int cpu, int flag) | |
2346 | { | |
2347 | struct task_struct *t = current; | |
2348 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2349 | |
c96d145e | 2350 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2351 | /* |
2352 | * If power savings logic is enabled for a domain, stop there. | |
2353 | */ | |
5c45bf27 SS |
2354 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2355 | break; | |
476d139c NP |
2356 | if (tmp->flags & flag) |
2357 | sd = tmp; | |
c96d145e | 2358 | } |
476d139c | 2359 | |
039a1c41 PZ |
2360 | if (sd) |
2361 | update_shares(sd); | |
2362 | ||
476d139c | 2363 | while (sd) { |
476d139c | 2364 | struct sched_group *group; |
1a848870 SS |
2365 | int new_cpu, weight; |
2366 | ||
2367 | if (!(sd->flags & flag)) { | |
2368 | sd = sd->child; | |
2369 | continue; | |
2370 | } | |
476d139c | 2371 | |
476d139c | 2372 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2373 | if (!group) { |
2374 | sd = sd->child; | |
2375 | continue; | |
2376 | } | |
476d139c | 2377 | |
758b2cdc | 2378 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2379 | if (new_cpu == -1 || new_cpu == cpu) { |
2380 | /* Now try balancing at a lower domain level of cpu */ | |
2381 | sd = sd->child; | |
2382 | continue; | |
2383 | } | |
476d139c | 2384 | |
1a848870 | 2385 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2386 | cpu = new_cpu; |
758b2cdc | 2387 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2388 | sd = NULL; |
476d139c | 2389 | for_each_domain(cpu, tmp) { |
758b2cdc | 2390 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2391 | break; |
2392 | if (tmp->flags & flag) | |
2393 | sd = tmp; | |
2394 | } | |
2395 | /* while loop will break here if sd == NULL */ | |
2396 | } | |
2397 | ||
2398 | return cpu; | |
2399 | } | |
2400 | ||
2401 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2402 | |
0793a61d TG |
2403 | /** |
2404 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2405 | * @p: the task to evaluate | |
2406 | * @func: the function to be called | |
2407 | * @info: the function call argument | |
2408 | * | |
2409 | * Calls the function @func when the task is currently running. This might | |
2410 | * be on the current CPU, which just calls the function directly | |
2411 | */ | |
2412 | void task_oncpu_function_call(struct task_struct *p, | |
2413 | void (*func) (void *info), void *info) | |
2414 | { | |
2415 | int cpu; | |
2416 | ||
2417 | preempt_disable(); | |
2418 | cpu = task_cpu(p); | |
2419 | if (task_curr(p)) | |
2420 | smp_call_function_single(cpu, func, info, 1); | |
2421 | preempt_enable(); | |
2422 | } | |
2423 | ||
1da177e4 LT |
2424 | /*** |
2425 | * try_to_wake_up - wake up a thread | |
2426 | * @p: the to-be-woken-up thread | |
2427 | * @state: the mask of task states that can be woken | |
2428 | * @sync: do a synchronous wakeup? | |
2429 | * | |
2430 | * Put it on the run-queue if it's not already there. The "current" | |
2431 | * thread is always on the run-queue (except when the actual | |
2432 | * re-schedule is in progress), and as such you're allowed to do | |
2433 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2434 | * runnable without the overhead of this. | |
2435 | * | |
2436 | * returns failure only if the task is already active. | |
2437 | */ | |
36c8b586 | 2438 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2439 | { |
cc367732 | 2440 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2441 | unsigned long flags; |
2442 | long old_state; | |
70b97a7f | 2443 | struct rq *rq; |
1da177e4 | 2444 | |
b85d0667 IM |
2445 | if (!sched_feat(SYNC_WAKEUPS)) |
2446 | sync = 0; | |
2447 | ||
2398f2c6 | 2448 | #ifdef CONFIG_SMP |
57310a98 | 2449 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2450 | struct sched_domain *sd; |
2451 | ||
2452 | this_cpu = raw_smp_processor_id(); | |
2453 | cpu = task_cpu(p); | |
2454 | ||
2455 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2456 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2457 | update_shares(sd); |
2458 | break; | |
2459 | } | |
2460 | } | |
2461 | } | |
2462 | #endif | |
2463 | ||
04e2f174 | 2464 | smp_wmb(); |
1da177e4 | 2465 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2466 | update_rq_clock(rq); |
1da177e4 LT |
2467 | old_state = p->state; |
2468 | if (!(old_state & state)) | |
2469 | goto out; | |
2470 | ||
dd41f596 | 2471 | if (p->se.on_rq) |
1da177e4 LT |
2472 | goto out_running; |
2473 | ||
2474 | cpu = task_cpu(p); | |
cc367732 | 2475 | orig_cpu = cpu; |
1da177e4 LT |
2476 | this_cpu = smp_processor_id(); |
2477 | ||
2478 | #ifdef CONFIG_SMP | |
2479 | if (unlikely(task_running(rq, p))) | |
2480 | goto out_activate; | |
2481 | ||
5d2f5a61 DA |
2482 | cpu = p->sched_class->select_task_rq(p, sync); |
2483 | if (cpu != orig_cpu) { | |
2484 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2485 | task_rq_unlock(rq, &flags); |
2486 | /* might preempt at this point */ | |
2487 | rq = task_rq_lock(p, &flags); | |
2488 | old_state = p->state; | |
2489 | if (!(old_state & state)) | |
2490 | goto out; | |
dd41f596 | 2491 | if (p->se.on_rq) |
1da177e4 LT |
2492 | goto out_running; |
2493 | ||
2494 | this_cpu = smp_processor_id(); | |
2495 | cpu = task_cpu(p); | |
2496 | } | |
2497 | ||
e7693a36 GH |
2498 | #ifdef CONFIG_SCHEDSTATS |
2499 | schedstat_inc(rq, ttwu_count); | |
2500 | if (cpu == this_cpu) | |
2501 | schedstat_inc(rq, ttwu_local); | |
2502 | else { | |
2503 | struct sched_domain *sd; | |
2504 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2505 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2506 | schedstat_inc(sd, ttwu_wake_remote); |
2507 | break; | |
2508 | } | |
2509 | } | |
2510 | } | |
6d6bc0ad | 2511 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2512 | |
1da177e4 LT |
2513 | out_activate: |
2514 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2515 | schedstat_inc(p, se.nr_wakeups); |
2516 | if (sync) | |
2517 | schedstat_inc(p, se.nr_wakeups_sync); | |
2518 | if (orig_cpu != cpu) | |
2519 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2520 | if (cpu == this_cpu) | |
2521 | schedstat_inc(p, se.nr_wakeups_local); | |
2522 | else | |
2523 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2524 | activate_task(rq, p, 1); |
1da177e4 LT |
2525 | success = 1; |
2526 | ||
831451ac PZ |
2527 | /* |
2528 | * Only attribute actual wakeups done by this task. | |
2529 | */ | |
2530 | if (!in_interrupt()) { | |
2531 | struct sched_entity *se = ¤t->se; | |
2532 | u64 sample = se->sum_exec_runtime; | |
2533 | ||
2534 | if (se->last_wakeup) | |
2535 | sample -= se->last_wakeup; | |
2536 | else | |
2537 | sample -= se->start_runtime; | |
2538 | update_avg(&se->avg_wakeup, sample); | |
2539 | ||
2540 | se->last_wakeup = se->sum_exec_runtime; | |
2541 | } | |
2542 | ||
1da177e4 | 2543 | out_running: |
468a15bb | 2544 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2545 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2546 | |
1da177e4 | 2547 | p->state = TASK_RUNNING; |
9a897c5a SR |
2548 | #ifdef CONFIG_SMP |
2549 | if (p->sched_class->task_wake_up) | |
2550 | p->sched_class->task_wake_up(rq, p); | |
2551 | #endif | |
1da177e4 LT |
2552 | out: |
2553 | task_rq_unlock(rq, &flags); | |
2554 | ||
2555 | return success; | |
2556 | } | |
2557 | ||
50fa610a DH |
2558 | /** |
2559 | * wake_up_process - Wake up a specific process | |
2560 | * @p: The process to be woken up. | |
2561 | * | |
2562 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2563 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2564 | * running. | |
2565 | * | |
2566 | * It may be assumed that this function implies a write memory barrier before | |
2567 | * changing the task state if and only if any tasks are woken up. | |
2568 | */ | |
7ad5b3a5 | 2569 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2570 | { |
d9514f6c | 2571 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2572 | } |
1da177e4 LT |
2573 | EXPORT_SYMBOL(wake_up_process); |
2574 | ||
7ad5b3a5 | 2575 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2576 | { |
2577 | return try_to_wake_up(p, state, 0); | |
2578 | } | |
2579 | ||
1da177e4 LT |
2580 | /* |
2581 | * Perform scheduler related setup for a newly forked process p. | |
2582 | * p is forked by current. | |
dd41f596 IM |
2583 | * |
2584 | * __sched_fork() is basic setup used by init_idle() too: | |
2585 | */ | |
2586 | static void __sched_fork(struct task_struct *p) | |
2587 | { | |
dd41f596 IM |
2588 | p->se.exec_start = 0; |
2589 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2590 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2591 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2592 | p->se.last_wakeup = 0; |
2593 | p->se.avg_overlap = 0; | |
831451ac PZ |
2594 | p->se.start_runtime = 0; |
2595 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2596 | |
2597 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2598 | p->se.wait_start = 0; |
2599 | p->se.wait_max = 0; | |
2600 | p->se.wait_count = 0; | |
2601 | p->se.wait_sum = 0; | |
2602 | ||
2603 | p->se.sleep_start = 0; | |
2604 | p->se.sleep_max = 0; | |
2605 | p->se.sum_sleep_runtime = 0; | |
2606 | ||
2607 | p->se.block_start = 0; | |
2608 | p->se.block_max = 0; | |
2609 | p->se.exec_max = 0; | |
2610 | p->se.slice_max = 0; | |
2611 | ||
2612 | p->se.nr_migrations_cold = 0; | |
2613 | p->se.nr_failed_migrations_affine = 0; | |
2614 | p->se.nr_failed_migrations_running = 0; | |
2615 | p->se.nr_failed_migrations_hot = 0; | |
2616 | p->se.nr_forced_migrations = 0; | |
2617 | p->se.nr_forced2_migrations = 0; | |
2618 | ||
2619 | p->se.nr_wakeups = 0; | |
2620 | p->se.nr_wakeups_sync = 0; | |
2621 | p->se.nr_wakeups_migrate = 0; | |
2622 | p->se.nr_wakeups_local = 0; | |
2623 | p->se.nr_wakeups_remote = 0; | |
2624 | p->se.nr_wakeups_affine = 0; | |
2625 | p->se.nr_wakeups_affine_attempts = 0; | |
2626 | p->se.nr_wakeups_passive = 0; | |
2627 | p->se.nr_wakeups_idle = 0; | |
2628 | ||
6cfb0d5d | 2629 | #endif |
476d139c | 2630 | |
fa717060 | 2631 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2632 | p->se.on_rq = 0; |
4a55bd5e | 2633 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2634 | |
e107be36 AK |
2635 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2636 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2637 | #endif | |
2638 | ||
1da177e4 LT |
2639 | /* |
2640 | * We mark the process as running here, but have not actually | |
2641 | * inserted it onto the runqueue yet. This guarantees that | |
2642 | * nobody will actually run it, and a signal or other external | |
2643 | * event cannot wake it up and insert it on the runqueue either. | |
2644 | */ | |
2645 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2646 | } |
2647 | ||
2648 | /* | |
2649 | * fork()/clone()-time setup: | |
2650 | */ | |
2651 | void sched_fork(struct task_struct *p, int clone_flags) | |
2652 | { | |
2653 | int cpu = get_cpu(); | |
2654 | ||
2655 | __sched_fork(p); | |
2656 | ||
2657 | #ifdef CONFIG_SMP | |
2658 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2659 | #endif | |
02e4bac2 | 2660 | set_task_cpu(p, cpu); |
b29739f9 IM |
2661 | |
2662 | /* | |
b9dc29e7 | 2663 | * Make sure we do not leak PI boosting priority to the child. |
b29739f9 | 2664 | */ |
b9dc29e7 | 2665 | p->prio = current->normal_prio; |
ca94c442 | 2666 | |
b9dc29e7 MG |
2667 | /* |
2668 | * Revert to default priority/policy on fork if requested. | |
2669 | */ | |
2670 | if (unlikely(p->sched_reset_on_fork)) { | |
2671 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) | |
2672 | p->policy = SCHED_NORMAL; | |
2673 | ||
2674 | if (p->normal_prio < DEFAULT_PRIO) | |
2675 | p->prio = DEFAULT_PRIO; | |
2676 | ||
6c697bdf MG |
2677 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2678 | p->static_prio = NICE_TO_PRIO(0); | |
2679 | set_load_weight(p); | |
2680 | } | |
2681 | ||
b9dc29e7 MG |
2682 | /* |
2683 | * We don't need the reset flag anymore after the fork. It has | |
2684 | * fulfilled its duty: | |
2685 | */ | |
2686 | p->sched_reset_on_fork = 0; | |
2687 | } | |
ca94c442 | 2688 | |
2ddbf952 HS |
2689 | if (!rt_prio(p->prio)) |
2690 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2691 | |
52f17b6c | 2692 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2693 | if (likely(sched_info_on())) |
52f17b6c | 2694 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2695 | #endif |
d6077cb8 | 2696 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2697 | p->oncpu = 0; |
2698 | #endif | |
1da177e4 | 2699 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2700 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2701 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2702 | #endif |
917b627d GH |
2703 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2704 | ||
476d139c | 2705 | put_cpu(); |
1da177e4 LT |
2706 | } |
2707 | ||
2708 | /* | |
2709 | * wake_up_new_task - wake up a newly created task for the first time. | |
2710 | * | |
2711 | * This function will do some initial scheduler statistics housekeeping | |
2712 | * that must be done for every newly created context, then puts the task | |
2713 | * on the runqueue and wakes it. | |
2714 | */ | |
7ad5b3a5 | 2715 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2716 | { |
2717 | unsigned long flags; | |
dd41f596 | 2718 | struct rq *rq; |
1da177e4 LT |
2719 | |
2720 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2721 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2722 | update_rq_clock(rq); |
1da177e4 LT |
2723 | |
2724 | p->prio = effective_prio(p); | |
2725 | ||
b9dca1e0 | 2726 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2727 | activate_task(rq, p, 0); |
1da177e4 | 2728 | } else { |
1da177e4 | 2729 | /* |
dd41f596 IM |
2730 | * Let the scheduling class do new task startup |
2731 | * management (if any): | |
1da177e4 | 2732 | */ |
ee0827d8 | 2733 | p->sched_class->task_new(rq, p); |
c09595f6 | 2734 | inc_nr_running(rq); |
1da177e4 | 2735 | } |
c71dd42d | 2736 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2737 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2738 | #ifdef CONFIG_SMP |
2739 | if (p->sched_class->task_wake_up) | |
2740 | p->sched_class->task_wake_up(rq, p); | |
2741 | #endif | |
dd41f596 | 2742 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2743 | } |
2744 | ||
e107be36 AK |
2745 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2746 | ||
2747 | /** | |
80dd99b3 | 2748 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2749 | * @notifier: notifier struct to register |
e107be36 AK |
2750 | */ |
2751 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2752 | { | |
2753 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2754 | } | |
2755 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2756 | ||
2757 | /** | |
2758 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2759 | * @notifier: notifier struct to unregister |
e107be36 AK |
2760 | * |
2761 | * This is safe to call from within a preemption notifier. | |
2762 | */ | |
2763 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2764 | { | |
2765 | hlist_del(¬ifier->link); | |
2766 | } | |
2767 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2768 | ||
2769 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2770 | { | |
2771 | struct preempt_notifier *notifier; | |
2772 | struct hlist_node *node; | |
2773 | ||
2774 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2775 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2776 | } | |
2777 | ||
2778 | static void | |
2779 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2780 | struct task_struct *next) | |
2781 | { | |
2782 | struct preempt_notifier *notifier; | |
2783 | struct hlist_node *node; | |
2784 | ||
2785 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2786 | notifier->ops->sched_out(notifier, next); | |
2787 | } | |
2788 | ||
6d6bc0ad | 2789 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2790 | |
2791 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2792 | { | |
2793 | } | |
2794 | ||
2795 | static void | |
2796 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2797 | struct task_struct *next) | |
2798 | { | |
2799 | } | |
2800 | ||
6d6bc0ad | 2801 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2802 | |
4866cde0 NP |
2803 | /** |
2804 | * prepare_task_switch - prepare to switch tasks | |
2805 | * @rq: the runqueue preparing to switch | |
421cee29 | 2806 | * @prev: the current task that is being switched out |
4866cde0 NP |
2807 | * @next: the task we are going to switch to. |
2808 | * | |
2809 | * This is called with the rq lock held and interrupts off. It must | |
2810 | * be paired with a subsequent finish_task_switch after the context | |
2811 | * switch. | |
2812 | * | |
2813 | * prepare_task_switch sets up locking and calls architecture specific | |
2814 | * hooks. | |
2815 | */ | |
e107be36 AK |
2816 | static inline void |
2817 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2818 | struct task_struct *next) | |
4866cde0 | 2819 | { |
e107be36 | 2820 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2821 | prepare_lock_switch(rq, next); |
2822 | prepare_arch_switch(next); | |
2823 | } | |
2824 | ||
1da177e4 LT |
2825 | /** |
2826 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2827 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2828 | * @prev: the thread we just switched away from. |
2829 | * | |
4866cde0 NP |
2830 | * finish_task_switch must be called after the context switch, paired |
2831 | * with a prepare_task_switch call before the context switch. | |
2832 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2833 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2834 | * |
2835 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2836 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2837 | * with the lock held can cause deadlocks; see schedule() for |
2838 | * details.) | |
2839 | */ | |
3f029d3c | 2840 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2841 | __releases(rq->lock) |
2842 | { | |
1da177e4 | 2843 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2844 | long prev_state; |
1da177e4 LT |
2845 | |
2846 | rq->prev_mm = NULL; | |
2847 | ||
2848 | /* | |
2849 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2850 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2851 | * schedule one last time. The schedule call will never return, and |
2852 | * the scheduled task must drop that reference. | |
c394cc9f | 2853 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2854 | * still held, otherwise prev could be scheduled on another cpu, die |
2855 | * there before we look at prev->state, and then the reference would | |
2856 | * be dropped twice. | |
2857 | * Manfred Spraul <manfred@colorfullife.com> | |
2858 | */ | |
55a101f8 | 2859 | prev_state = prev->state; |
4866cde0 | 2860 | finish_arch_switch(prev); |
0793a61d | 2861 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2862 | finish_lock_switch(rq, prev); |
e8fa1362 | 2863 | |
e107be36 | 2864 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2865 | if (mm) |
2866 | mmdrop(mm); | |
c394cc9f | 2867 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2868 | /* |
2869 | * Remove function-return probe instances associated with this | |
2870 | * task and put them back on the free list. | |
9761eea8 | 2871 | */ |
c6fd91f0 | 2872 | kprobe_flush_task(prev); |
1da177e4 | 2873 | put_task_struct(prev); |
c6fd91f0 | 2874 | } |
3f029d3c GH |
2875 | } |
2876 | ||
2877 | #ifdef CONFIG_SMP | |
2878 | ||
2879 | /* assumes rq->lock is held */ | |
2880 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2881 | { | |
2882 | if (prev->sched_class->pre_schedule) | |
2883 | prev->sched_class->pre_schedule(rq, prev); | |
2884 | } | |
2885 | ||
2886 | /* rq->lock is NOT held, but preemption is disabled */ | |
2887 | static inline void post_schedule(struct rq *rq) | |
2888 | { | |
2889 | if (rq->post_schedule) { | |
2890 | unsigned long flags; | |
2891 | ||
2892 | spin_lock_irqsave(&rq->lock, flags); | |
2893 | if (rq->curr->sched_class->post_schedule) | |
2894 | rq->curr->sched_class->post_schedule(rq); | |
2895 | spin_unlock_irqrestore(&rq->lock, flags); | |
2896 | ||
2897 | rq->post_schedule = 0; | |
2898 | } | |
2899 | } | |
2900 | ||
2901 | #else | |
da19ab51 | 2902 | |
3f029d3c GH |
2903 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2904 | { | |
2905 | } | |
2906 | ||
2907 | static inline void post_schedule(struct rq *rq) | |
2908 | { | |
1da177e4 LT |
2909 | } |
2910 | ||
3f029d3c GH |
2911 | #endif |
2912 | ||
1da177e4 LT |
2913 | /** |
2914 | * schedule_tail - first thing a freshly forked thread must call. | |
2915 | * @prev: the thread we just switched away from. | |
2916 | */ | |
36c8b586 | 2917 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2918 | __releases(rq->lock) |
2919 | { | |
70b97a7f | 2920 | struct rq *rq = this_rq(); |
da19ab51 | 2921 | |
3f029d3c | 2922 | finish_task_switch(rq, prev); |
da19ab51 | 2923 | |
3f029d3c GH |
2924 | /* |
2925 | * FIXME: do we need to worry about rq being invalidated by the | |
2926 | * task_switch? | |
2927 | */ | |
2928 | post_schedule(rq); | |
70b97a7f | 2929 | |
4866cde0 NP |
2930 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2931 | /* In this case, finish_task_switch does not reenable preemption */ | |
2932 | preempt_enable(); | |
2933 | #endif | |
1da177e4 | 2934 | if (current->set_child_tid) |
b488893a | 2935 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2936 | } |
2937 | ||
2938 | /* | |
2939 | * context_switch - switch to the new MM and the new | |
2940 | * thread's register state. | |
2941 | */ | |
3f029d3c | 2942 | static inline void |
70b97a7f | 2943 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2944 | struct task_struct *next) |
1da177e4 | 2945 | { |
dd41f596 | 2946 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2947 | |
e107be36 | 2948 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2949 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2950 | mm = next->mm; |
2951 | oldmm = prev->active_mm; | |
9226d125 ZA |
2952 | /* |
2953 | * For paravirt, this is coupled with an exit in switch_to to | |
2954 | * combine the page table reload and the switch backend into | |
2955 | * one hypercall. | |
2956 | */ | |
224101ed | 2957 | arch_start_context_switch(prev); |
9226d125 | 2958 | |
dd41f596 | 2959 | if (unlikely(!mm)) { |
1da177e4 LT |
2960 | next->active_mm = oldmm; |
2961 | atomic_inc(&oldmm->mm_count); | |
2962 | enter_lazy_tlb(oldmm, next); | |
2963 | } else | |
2964 | switch_mm(oldmm, mm, next); | |
2965 | ||
dd41f596 | 2966 | if (unlikely(!prev->mm)) { |
1da177e4 | 2967 | prev->active_mm = NULL; |
1da177e4 LT |
2968 | rq->prev_mm = oldmm; |
2969 | } | |
3a5f5e48 IM |
2970 | /* |
2971 | * Since the runqueue lock will be released by the next | |
2972 | * task (which is an invalid locking op but in the case | |
2973 | * of the scheduler it's an obvious special-case), so we | |
2974 | * do an early lockdep release here: | |
2975 | */ | |
2976 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2977 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2978 | #endif |
1da177e4 LT |
2979 | |
2980 | /* Here we just switch the register state and the stack. */ | |
2981 | switch_to(prev, next, prev); | |
2982 | ||
dd41f596 IM |
2983 | barrier(); |
2984 | /* | |
2985 | * this_rq must be evaluated again because prev may have moved | |
2986 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2987 | * frame will be invalid. | |
2988 | */ | |
3f029d3c | 2989 | finish_task_switch(this_rq(), prev); |
1da177e4 LT |
2990 | } |
2991 | ||
2992 | /* | |
2993 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2994 | * | |
2995 | * externally visible scheduler statistics: current number of runnable | |
2996 | * threads, current number of uninterruptible-sleeping threads, total | |
2997 | * number of context switches performed since bootup. | |
2998 | */ | |
2999 | unsigned long nr_running(void) | |
3000 | { | |
3001 | unsigned long i, sum = 0; | |
3002 | ||
3003 | for_each_online_cpu(i) | |
3004 | sum += cpu_rq(i)->nr_running; | |
3005 | ||
3006 | return sum; | |
3007 | } | |
3008 | ||
3009 | unsigned long nr_uninterruptible(void) | |
3010 | { | |
3011 | unsigned long i, sum = 0; | |
3012 | ||
0a945022 | 3013 | for_each_possible_cpu(i) |
1da177e4 LT |
3014 | sum += cpu_rq(i)->nr_uninterruptible; |
3015 | ||
3016 | /* | |
3017 | * Since we read the counters lockless, it might be slightly | |
3018 | * inaccurate. Do not allow it to go below zero though: | |
3019 | */ | |
3020 | if (unlikely((long)sum < 0)) | |
3021 | sum = 0; | |
3022 | ||
3023 | return sum; | |
3024 | } | |
3025 | ||
3026 | unsigned long long nr_context_switches(void) | |
3027 | { | |
cc94abfc SR |
3028 | int i; |
3029 | unsigned long long sum = 0; | |
1da177e4 | 3030 | |
0a945022 | 3031 | for_each_possible_cpu(i) |
1da177e4 LT |
3032 | sum += cpu_rq(i)->nr_switches; |
3033 | ||
3034 | return sum; | |
3035 | } | |
3036 | ||
3037 | unsigned long nr_iowait(void) | |
3038 | { | |
3039 | unsigned long i, sum = 0; | |
3040 | ||
0a945022 | 3041 | for_each_possible_cpu(i) |
1da177e4 LT |
3042 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
3043 | ||
3044 | return sum; | |
3045 | } | |
3046 | ||
dce48a84 TG |
3047 | /* Variables and functions for calc_load */ |
3048 | static atomic_long_t calc_load_tasks; | |
3049 | static unsigned long calc_load_update; | |
3050 | unsigned long avenrun[3]; | |
3051 | EXPORT_SYMBOL(avenrun); | |
3052 | ||
2d02494f TG |
3053 | /** |
3054 | * get_avenrun - get the load average array | |
3055 | * @loads: pointer to dest load array | |
3056 | * @offset: offset to add | |
3057 | * @shift: shift count to shift the result left | |
3058 | * | |
3059 | * These values are estimates at best, so no need for locking. | |
3060 | */ | |
3061 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3062 | { | |
3063 | loads[0] = (avenrun[0] + offset) << shift; | |
3064 | loads[1] = (avenrun[1] + offset) << shift; | |
3065 | loads[2] = (avenrun[2] + offset) << shift; | |
3066 | } | |
3067 | ||
dce48a84 TG |
3068 | static unsigned long |
3069 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3070 | { |
dce48a84 TG |
3071 | load *= exp; |
3072 | load += active * (FIXED_1 - exp); | |
3073 | return load >> FSHIFT; | |
3074 | } | |
db1b1fef | 3075 | |
dce48a84 TG |
3076 | /* |
3077 | * calc_load - update the avenrun load estimates 10 ticks after the | |
3078 | * CPUs have updated calc_load_tasks. | |
3079 | */ | |
3080 | void calc_global_load(void) | |
3081 | { | |
3082 | unsigned long upd = calc_load_update + 10; | |
3083 | long active; | |
3084 | ||
3085 | if (time_before(jiffies, upd)) | |
3086 | return; | |
db1b1fef | 3087 | |
dce48a84 TG |
3088 | active = atomic_long_read(&calc_load_tasks); |
3089 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 3090 | |
dce48a84 TG |
3091 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3092 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3093 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
3094 | ||
3095 | calc_load_update += LOAD_FREQ; | |
3096 | } | |
3097 | ||
3098 | /* | |
3099 | * Either called from update_cpu_load() or from a cpu going idle | |
3100 | */ | |
3101 | static void calc_load_account_active(struct rq *this_rq) | |
3102 | { | |
3103 | long nr_active, delta; | |
3104 | ||
3105 | nr_active = this_rq->nr_running; | |
3106 | nr_active += (long) this_rq->nr_uninterruptible; | |
3107 | ||
3108 | if (nr_active != this_rq->calc_load_active) { | |
3109 | delta = nr_active - this_rq->calc_load_active; | |
3110 | this_rq->calc_load_active = nr_active; | |
3111 | atomic_long_add(delta, &calc_load_tasks); | |
3112 | } | |
db1b1fef JS |
3113 | } |
3114 | ||
23a185ca PM |
3115 | /* |
3116 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
3117 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
3118 | */ | |
23a185ca PM |
3119 | u64 cpu_nr_migrations(int cpu) |
3120 | { | |
3121 | return cpu_rq(cpu)->nr_migrations_in; | |
3122 | } | |
3123 | ||
48f24c4d | 3124 | /* |
dd41f596 IM |
3125 | * Update rq->cpu_load[] statistics. This function is usually called every |
3126 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3127 | */ |
dd41f596 | 3128 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3129 | { |
495eca49 | 3130 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3131 | int i, scale; |
3132 | ||
3133 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3134 | |
3135 | /* Update our load: */ | |
3136 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3137 | unsigned long old_load, new_load; | |
3138 | ||
3139 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3140 | ||
3141 | old_load = this_rq->cpu_load[i]; | |
3142 | new_load = this_load; | |
a25707f3 IM |
3143 | /* |
3144 | * Round up the averaging division if load is increasing. This | |
3145 | * prevents us from getting stuck on 9 if the load is 10, for | |
3146 | * example. | |
3147 | */ | |
3148 | if (new_load > old_load) | |
3149 | new_load += scale-1; | |
dd41f596 IM |
3150 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3151 | } | |
dce48a84 TG |
3152 | |
3153 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3154 | this_rq->calc_load_update += LOAD_FREQ; | |
3155 | calc_load_account_active(this_rq); | |
3156 | } | |
48f24c4d IM |
3157 | } |
3158 | ||
dd41f596 IM |
3159 | #ifdef CONFIG_SMP |
3160 | ||
1da177e4 LT |
3161 | /* |
3162 | * double_rq_lock - safely lock two runqueues | |
3163 | * | |
3164 | * Note this does not disable interrupts like task_rq_lock, | |
3165 | * you need to do so manually before calling. | |
3166 | */ | |
70b97a7f | 3167 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3168 | __acquires(rq1->lock) |
3169 | __acquires(rq2->lock) | |
3170 | { | |
054b9108 | 3171 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3172 | if (rq1 == rq2) { |
3173 | spin_lock(&rq1->lock); | |
3174 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3175 | } else { | |
c96d145e | 3176 | if (rq1 < rq2) { |
1da177e4 | 3177 | spin_lock(&rq1->lock); |
5e710e37 | 3178 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3179 | } else { |
3180 | spin_lock(&rq2->lock); | |
5e710e37 | 3181 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3182 | } |
3183 | } | |
6e82a3be IM |
3184 | update_rq_clock(rq1); |
3185 | update_rq_clock(rq2); | |
1da177e4 LT |
3186 | } |
3187 | ||
3188 | /* | |
3189 | * double_rq_unlock - safely unlock two runqueues | |
3190 | * | |
3191 | * Note this does not restore interrupts like task_rq_unlock, | |
3192 | * you need to do so manually after calling. | |
3193 | */ | |
70b97a7f | 3194 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3195 | __releases(rq1->lock) |
3196 | __releases(rq2->lock) | |
3197 | { | |
3198 | spin_unlock(&rq1->lock); | |
3199 | if (rq1 != rq2) | |
3200 | spin_unlock(&rq2->lock); | |
3201 | else | |
3202 | __release(rq2->lock); | |
3203 | } | |
3204 | ||
1da177e4 LT |
3205 | /* |
3206 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3207 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3208 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3209 | * the cpu_allowed mask is restored. |
3210 | */ | |
36c8b586 | 3211 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3212 | { |
70b97a7f | 3213 | struct migration_req req; |
1da177e4 | 3214 | unsigned long flags; |
70b97a7f | 3215 | struct rq *rq; |
1da177e4 LT |
3216 | |
3217 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3218 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3219 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3220 | goto out; |
3221 | ||
3222 | /* force the process onto the specified CPU */ | |
3223 | if (migrate_task(p, dest_cpu, &req)) { | |
3224 | /* Need to wait for migration thread (might exit: take ref). */ | |
3225 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3226 | |
1da177e4 LT |
3227 | get_task_struct(mt); |
3228 | task_rq_unlock(rq, &flags); | |
3229 | wake_up_process(mt); | |
3230 | put_task_struct(mt); | |
3231 | wait_for_completion(&req.done); | |
36c8b586 | 3232 | |
1da177e4 LT |
3233 | return; |
3234 | } | |
3235 | out: | |
3236 | task_rq_unlock(rq, &flags); | |
3237 | } | |
3238 | ||
3239 | /* | |
476d139c NP |
3240 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3241 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3242 | */ |
3243 | void sched_exec(void) | |
3244 | { | |
1da177e4 | 3245 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3246 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3247 | put_cpu(); |
476d139c NP |
3248 | if (new_cpu != this_cpu) |
3249 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3250 | } |
3251 | ||
3252 | /* | |
3253 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3254 | * Both runqueues must be locked. | |
3255 | */ | |
dd41f596 IM |
3256 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3257 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3258 | { |
2e1cb74a | 3259 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3260 | set_task_cpu(p, this_cpu); |
dd41f596 | 3261 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3262 | /* |
3263 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3264 | * to be always true for them. | |
3265 | */ | |
15afe09b | 3266 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3267 | } |
3268 | ||
3269 | /* | |
3270 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3271 | */ | |
858119e1 | 3272 | static |
70b97a7f | 3273 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3274 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3275 | int *all_pinned) |
1da177e4 | 3276 | { |
708dc512 | 3277 | int tsk_cache_hot = 0; |
1da177e4 LT |
3278 | /* |
3279 | * We do not migrate tasks that are: | |
3280 | * 1) running (obviously), or | |
3281 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3282 | * 3) are cache-hot on their current CPU. | |
3283 | */ | |
96f874e2 | 3284 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3285 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3286 | return 0; |
cc367732 | 3287 | } |
81026794 NP |
3288 | *all_pinned = 0; |
3289 | ||
cc367732 IM |
3290 | if (task_running(rq, p)) { |
3291 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3292 | return 0; |
cc367732 | 3293 | } |
1da177e4 | 3294 | |
da84d961 IM |
3295 | /* |
3296 | * Aggressive migration if: | |
3297 | * 1) task is cache cold, or | |
3298 | * 2) too many balance attempts have failed. | |
3299 | */ | |
3300 | ||
708dc512 LH |
3301 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3302 | if (!tsk_cache_hot || | |
3303 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3304 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3305 | if (tsk_cache_hot) { |
da84d961 | 3306 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3307 | schedstat_inc(p, se.nr_forced_migrations); |
3308 | } | |
da84d961 IM |
3309 | #endif |
3310 | return 1; | |
3311 | } | |
3312 | ||
708dc512 | 3313 | if (tsk_cache_hot) { |
cc367732 | 3314 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3315 | return 0; |
cc367732 | 3316 | } |
1da177e4 LT |
3317 | return 1; |
3318 | } | |
3319 | ||
e1d1484f PW |
3320 | static unsigned long |
3321 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3322 | unsigned long max_load_move, struct sched_domain *sd, | |
3323 | enum cpu_idle_type idle, int *all_pinned, | |
3324 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3325 | { |
051c6764 | 3326 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3327 | struct task_struct *p; |
3328 | long rem_load_move = max_load_move; | |
1da177e4 | 3329 | |
e1d1484f | 3330 | if (max_load_move == 0) |
1da177e4 LT |
3331 | goto out; |
3332 | ||
81026794 NP |
3333 | pinned = 1; |
3334 | ||
1da177e4 | 3335 | /* |
dd41f596 | 3336 | * Start the load-balancing iterator: |
1da177e4 | 3337 | */ |
dd41f596 IM |
3338 | p = iterator->start(iterator->arg); |
3339 | next: | |
b82d9fdd | 3340 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3341 | goto out; |
051c6764 PZ |
3342 | |
3343 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3344 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3345 | p = iterator->next(iterator->arg); |
3346 | goto next; | |
1da177e4 LT |
3347 | } |
3348 | ||
dd41f596 | 3349 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3350 | pulled++; |
dd41f596 | 3351 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3352 | |
7e96fa58 GH |
3353 | #ifdef CONFIG_PREEMPT |
3354 | /* | |
3355 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3356 | * will stop after the first task is pulled to minimize the critical | |
3357 | * section. | |
3358 | */ | |
3359 | if (idle == CPU_NEWLY_IDLE) | |
3360 | goto out; | |
3361 | #endif | |
3362 | ||
2dd73a4f | 3363 | /* |
b82d9fdd | 3364 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3365 | */ |
e1d1484f | 3366 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3367 | if (p->prio < *this_best_prio) |
3368 | *this_best_prio = p->prio; | |
dd41f596 IM |
3369 | p = iterator->next(iterator->arg); |
3370 | goto next; | |
1da177e4 LT |
3371 | } |
3372 | out: | |
3373 | /* | |
e1d1484f | 3374 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3375 | * so we can safely collect pull_task() stats here rather than |
3376 | * inside pull_task(). | |
3377 | */ | |
3378 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3379 | |
3380 | if (all_pinned) | |
3381 | *all_pinned = pinned; | |
e1d1484f PW |
3382 | |
3383 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3384 | } |
3385 | ||
dd41f596 | 3386 | /* |
43010659 PW |
3387 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3388 | * this_rq, as part of a balancing operation within domain "sd". | |
3389 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3390 | * |
3391 | * Called with both runqueues locked. | |
3392 | */ | |
3393 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3394 | unsigned long max_load_move, |
dd41f596 IM |
3395 | struct sched_domain *sd, enum cpu_idle_type idle, |
3396 | int *all_pinned) | |
3397 | { | |
5522d5d5 | 3398 | const struct sched_class *class = sched_class_highest; |
43010659 | 3399 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3400 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3401 | |
3402 | do { | |
43010659 PW |
3403 | total_load_moved += |
3404 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3405 | max_load_move - total_load_moved, |
a4ac01c3 | 3406 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3407 | class = class->next; |
c4acb2c0 | 3408 | |
7e96fa58 GH |
3409 | #ifdef CONFIG_PREEMPT |
3410 | /* | |
3411 | * NEWIDLE balancing is a source of latency, so preemptible | |
3412 | * kernels will stop after the first task is pulled to minimize | |
3413 | * the critical section. | |
3414 | */ | |
c4acb2c0 GH |
3415 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3416 | break; | |
7e96fa58 | 3417 | #endif |
43010659 | 3418 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3419 | |
43010659 PW |
3420 | return total_load_moved > 0; |
3421 | } | |
3422 | ||
e1d1484f PW |
3423 | static int |
3424 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3425 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3426 | struct rq_iterator *iterator) | |
3427 | { | |
3428 | struct task_struct *p = iterator->start(iterator->arg); | |
3429 | int pinned = 0; | |
3430 | ||
3431 | while (p) { | |
3432 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3433 | pull_task(busiest, p, this_rq, this_cpu); | |
3434 | /* | |
3435 | * Right now, this is only the second place pull_task() | |
3436 | * is called, so we can safely collect pull_task() | |
3437 | * stats here rather than inside pull_task(). | |
3438 | */ | |
3439 | schedstat_inc(sd, lb_gained[idle]); | |
3440 | ||
3441 | return 1; | |
3442 | } | |
3443 | p = iterator->next(iterator->arg); | |
3444 | } | |
3445 | ||
3446 | return 0; | |
3447 | } | |
3448 | ||
43010659 PW |
3449 | /* |
3450 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3451 | * part of active balancing operations within "domain". | |
3452 | * Returns 1 if successful and 0 otherwise. | |
3453 | * | |
3454 | * Called with both runqueues locked. | |
3455 | */ | |
3456 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3457 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3458 | { | |
5522d5d5 | 3459 | const struct sched_class *class; |
43010659 | 3460 | |
cde7e5ca | 3461 | for_each_class(class) { |
e1d1484f | 3462 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 | 3463 | return 1; |
cde7e5ca | 3464 | } |
43010659 PW |
3465 | |
3466 | return 0; | |
dd41f596 | 3467 | } |
67bb6c03 | 3468 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3469 | /* |
222d656d GS |
3470 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3471 | * during load balancing. | |
1da177e4 | 3472 | */ |
222d656d GS |
3473 | struct sd_lb_stats { |
3474 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3475 | struct sched_group *this; /* Local group in this sd */ | |
3476 | unsigned long total_load; /* Total load of all groups in sd */ | |
3477 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3478 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3479 | ||
3480 | /** Statistics of this group */ | |
3481 | unsigned long this_load; | |
3482 | unsigned long this_load_per_task; | |
3483 | unsigned long this_nr_running; | |
3484 | ||
3485 | /* Statistics of the busiest group */ | |
3486 | unsigned long max_load; | |
3487 | unsigned long busiest_load_per_task; | |
3488 | unsigned long busiest_nr_running; | |
3489 | ||
3490 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3491 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3492 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3493 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3494 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3495 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3496 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3497 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3498 | #endif |
222d656d | 3499 | }; |
1da177e4 | 3500 | |
d5ac537e | 3501 | /* |
381be78f GS |
3502 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3503 | */ | |
3504 | struct sg_lb_stats { | |
3505 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3506 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3507 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3508 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3509 | unsigned long group_capacity; | |
3510 | int group_imb; /* Is there an imbalance in the group ? */ | |
3511 | }; | |
408ed066 | 3512 | |
67bb6c03 GS |
3513 | /** |
3514 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3515 | * @group: The group whose first cpu is to be returned. | |
3516 | */ | |
3517 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3518 | { | |
3519 | return cpumask_first(sched_group_cpus(group)); | |
3520 | } | |
3521 | ||
3522 | /** | |
3523 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3524 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3525 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3526 | */ | |
3527 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3528 | enum cpu_idle_type idle) | |
3529 | { | |
3530 | int load_idx; | |
3531 | ||
3532 | switch (idle) { | |
3533 | case CPU_NOT_IDLE: | |
7897986b | 3534 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3535 | break; |
3536 | ||
3537 | case CPU_NEWLY_IDLE: | |
7897986b | 3538 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3539 | break; |
3540 | default: | |
7897986b | 3541 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3542 | break; |
3543 | } | |
1da177e4 | 3544 | |
67bb6c03 GS |
3545 | return load_idx; |
3546 | } | |
1da177e4 | 3547 | |
1da177e4 | 3548 | |
c071df18 GS |
3549 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3550 | /** | |
3551 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3552 | * the given sched_domain, during load balancing. | |
3553 | * | |
3554 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3555 | * @sds: Variable containing the statistics for sd. | |
3556 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3557 | */ | |
3558 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3559 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3560 | { | |
3561 | /* | |
3562 | * Busy processors will not participate in power savings | |
3563 | * balance. | |
3564 | */ | |
3565 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3566 | sds->power_savings_balance = 0; | |
3567 | else { | |
3568 | sds->power_savings_balance = 1; | |
3569 | sds->min_nr_running = ULONG_MAX; | |
3570 | sds->leader_nr_running = 0; | |
3571 | } | |
3572 | } | |
783609c6 | 3573 | |
c071df18 GS |
3574 | /** |
3575 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3576 | * sched_domain while performing load balancing. | |
3577 | * | |
3578 | * @group: sched_group belonging to the sched_domain under consideration. | |
3579 | * @sds: Variable containing the statistics of the sched_domain | |
3580 | * @local_group: Does group contain the CPU for which we're performing | |
3581 | * load balancing ? | |
3582 | * @sgs: Variable containing the statistics of the group. | |
3583 | */ | |
3584 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3585 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3586 | { | |
408ed066 | 3587 | |
c071df18 GS |
3588 | if (!sds->power_savings_balance) |
3589 | return; | |
1da177e4 | 3590 | |
c071df18 GS |
3591 | /* |
3592 | * If the local group is idle or completely loaded | |
3593 | * no need to do power savings balance at this domain | |
3594 | */ | |
3595 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3596 | !sds->this_nr_running)) | |
3597 | sds->power_savings_balance = 0; | |
2dd73a4f | 3598 | |
c071df18 GS |
3599 | /* |
3600 | * If a group is already running at full capacity or idle, | |
3601 | * don't include that group in power savings calculations | |
3602 | */ | |
3603 | if (!sds->power_savings_balance || | |
3604 | sgs->sum_nr_running >= sgs->group_capacity || | |
3605 | !sgs->sum_nr_running) | |
3606 | return; | |
5969fe06 | 3607 | |
c071df18 GS |
3608 | /* |
3609 | * Calculate the group which has the least non-idle load. | |
3610 | * This is the group from where we need to pick up the load | |
3611 | * for saving power | |
3612 | */ | |
3613 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3614 | (sgs->sum_nr_running == sds->min_nr_running && | |
3615 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3616 | sds->group_min = group; | |
3617 | sds->min_nr_running = sgs->sum_nr_running; | |
3618 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3619 | sgs->sum_nr_running; | |
3620 | } | |
783609c6 | 3621 | |
c071df18 GS |
3622 | /* |
3623 | * Calculate the group which is almost near its | |
3624 | * capacity but still has some space to pick up some load | |
3625 | * from other group and save more power | |
3626 | */ | |
3627 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3628 | return; | |
1da177e4 | 3629 | |
c071df18 GS |
3630 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3631 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3632 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3633 | sds->group_leader = group; | |
3634 | sds->leader_nr_running = sgs->sum_nr_running; | |
3635 | } | |
3636 | } | |
408ed066 | 3637 | |
c071df18 | 3638 | /** |
d5ac537e | 3639 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3640 | * @sds: Variable containing the statistics of the sched_domain |
3641 | * under consideration. | |
3642 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3643 | * @imbalance: Variable to store the imbalance. | |
3644 | * | |
d5ac537e RD |
3645 | * Description: |
3646 | * Check if we have potential to perform some power-savings balance. | |
3647 | * If yes, set the busiest group to be the least loaded group in the | |
3648 | * sched_domain, so that it's CPUs can be put to idle. | |
3649 | * | |
c071df18 GS |
3650 | * Returns 1 if there is potential to perform power-savings balance. |
3651 | * Else returns 0. | |
3652 | */ | |
3653 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3654 | int this_cpu, unsigned long *imbalance) | |
3655 | { | |
3656 | if (!sds->power_savings_balance) | |
3657 | return 0; | |
1da177e4 | 3658 | |
c071df18 GS |
3659 | if (sds->this != sds->group_leader || |
3660 | sds->group_leader == sds->group_min) | |
3661 | return 0; | |
783609c6 | 3662 | |
c071df18 GS |
3663 | *imbalance = sds->min_load_per_task; |
3664 | sds->busiest = sds->group_min; | |
1da177e4 | 3665 | |
c071df18 GS |
3666 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3667 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3668 | group_first_cpu(sds->group_leader); | |
3669 | } | |
3670 | ||
3671 | return 1; | |
1da177e4 | 3672 | |
c071df18 GS |
3673 | } |
3674 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3675 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3676 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3677 | { | |
3678 | return; | |
3679 | } | |
408ed066 | 3680 | |
c071df18 GS |
3681 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3682 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3683 | { | |
3684 | return; | |
3685 | } | |
3686 | ||
3687 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3688 | int this_cpu, unsigned long *imbalance) | |
3689 | { | |
3690 | return 0; | |
3691 | } | |
3692 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3693 | ||
3694 | ||
1f8c553d GS |
3695 | /** |
3696 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3697 | * @group: sched_group whose statistics are to be updated. | |
3698 | * @this_cpu: Cpu for which load balance is currently performed. | |
3699 | * @idle: Idle status of this_cpu | |
3700 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3701 | * @sd_idle: Idle status of the sched_domain containing group. | |
3702 | * @local_group: Does group contain this_cpu. | |
3703 | * @cpus: Set of cpus considered for load balancing. | |
3704 | * @balance: Should we balance. | |
3705 | * @sgs: variable to hold the statistics for this group. | |
3706 | */ | |
3707 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3708 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3709 | int local_group, const struct cpumask *cpus, | |
3710 | int *balance, struct sg_lb_stats *sgs) | |
3711 | { | |
3712 | unsigned long load, max_cpu_load, min_cpu_load; | |
3713 | int i; | |
3714 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3715 | unsigned long sum_avg_load_per_task; | |
3716 | unsigned long avg_load_per_task; | |
3717 | ||
3718 | if (local_group) | |
3719 | balance_cpu = group_first_cpu(group); | |
3720 | ||
3721 | /* Tally up the load of all CPUs in the group */ | |
3722 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3723 | max_cpu_load = 0; | |
3724 | min_cpu_load = ~0UL; | |
408ed066 | 3725 | |
1f8c553d GS |
3726 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3727 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3728 | |
1f8c553d GS |
3729 | if (*sd_idle && rq->nr_running) |
3730 | *sd_idle = 0; | |
5c45bf27 | 3731 | |
1f8c553d | 3732 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3733 | if (local_group) { |
1f8c553d GS |
3734 | if (idle_cpu(i) && !first_idle_cpu) { |
3735 | first_idle_cpu = 1; | |
3736 | balance_cpu = i; | |
3737 | } | |
3738 | ||
3739 | load = target_load(i, load_idx); | |
3740 | } else { | |
3741 | load = source_load(i, load_idx); | |
3742 | if (load > max_cpu_load) | |
3743 | max_cpu_load = load; | |
3744 | if (min_cpu_load > load) | |
3745 | min_cpu_load = load; | |
1da177e4 | 3746 | } |
5c45bf27 | 3747 | |
1f8c553d GS |
3748 | sgs->group_load += load; |
3749 | sgs->sum_nr_running += rq->nr_running; | |
3750 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3751 | |
1f8c553d GS |
3752 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3753 | } | |
5c45bf27 | 3754 | |
1f8c553d GS |
3755 | /* |
3756 | * First idle cpu or the first cpu(busiest) in this sched group | |
3757 | * is eligible for doing load balancing at this and above | |
3758 | * domains. In the newly idle case, we will allow all the cpu's | |
3759 | * to do the newly idle load balance. | |
3760 | */ | |
3761 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3762 | balance_cpu != this_cpu && balance) { | |
3763 | *balance = 0; | |
3764 | return; | |
3765 | } | |
5c45bf27 | 3766 | |
1f8c553d GS |
3767 | /* Adjust by relative CPU power of the group */ |
3768 | sgs->avg_load = sg_div_cpu_power(group, | |
3769 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3770 | |
1f8c553d GS |
3771 | |
3772 | /* | |
3773 | * Consider the group unbalanced when the imbalance is larger | |
3774 | * than the average weight of two tasks. | |
3775 | * | |
3776 | * APZ: with cgroup the avg task weight can vary wildly and | |
3777 | * might not be a suitable number - should we keep a | |
3778 | * normalized nr_running number somewhere that negates | |
3779 | * the hierarchy? | |
3780 | */ | |
3781 | avg_load_per_task = sg_div_cpu_power(group, | |
3782 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3783 | ||
3784 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3785 | sgs->group_imb = 1; | |
3786 | ||
3787 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3788 | ||
3789 | } | |
dd41f596 | 3790 | |
37abe198 GS |
3791 | /** |
3792 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3793 | * @sd: sched_domain whose statistics are to be updated. | |
3794 | * @this_cpu: Cpu for which load balance is currently performed. | |
3795 | * @idle: Idle status of this_cpu | |
3796 | * @sd_idle: Idle status of the sched_domain containing group. | |
3797 | * @cpus: Set of cpus considered for load balancing. | |
3798 | * @balance: Should we balance. | |
3799 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3800 | */ |
37abe198 GS |
3801 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3802 | enum cpu_idle_type idle, int *sd_idle, | |
3803 | const struct cpumask *cpus, int *balance, | |
3804 | struct sd_lb_stats *sds) | |
1da177e4 | 3805 | { |
222d656d | 3806 | struct sched_group *group = sd->groups; |
37abe198 | 3807 | struct sg_lb_stats sgs; |
222d656d GS |
3808 | int load_idx; |
3809 | ||
c071df18 | 3810 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3811 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3812 | |
3813 | do { | |
1da177e4 | 3814 | int local_group; |
1da177e4 | 3815 | |
758b2cdc RR |
3816 | local_group = cpumask_test_cpu(this_cpu, |
3817 | sched_group_cpus(group)); | |
381be78f | 3818 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3819 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3820 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3821 | |
37abe198 GS |
3822 | if (local_group && balance && !(*balance)) |
3823 | return; | |
783609c6 | 3824 | |
37abe198 GS |
3825 | sds->total_load += sgs.group_load; |
3826 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3827 | |
1da177e4 | 3828 | if (local_group) { |
37abe198 GS |
3829 | sds->this_load = sgs.avg_load; |
3830 | sds->this = group; | |
3831 | sds->this_nr_running = sgs.sum_nr_running; | |
3832 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3833 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3834 | (sgs.sum_nr_running > sgs.group_capacity || |
3835 | sgs.group_imb)) { | |
37abe198 GS |
3836 | sds->max_load = sgs.avg_load; |
3837 | sds->busiest = group; | |
3838 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3839 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3840 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3841 | } |
5c45bf27 | 3842 | |
c071df18 | 3843 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3844 | group = group->next; |
3845 | } while (group != sd->groups); | |
3846 | ||
37abe198 | 3847 | } |
1da177e4 | 3848 | |
2e6f44ae GS |
3849 | /** |
3850 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3851 | * amongst the groups of a sched_domain, during |
3852 | * load balancing. | |
2e6f44ae GS |
3853 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3854 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3855 | * @imbalance: Variable to store the imbalance. | |
3856 | */ | |
3857 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3858 | int this_cpu, unsigned long *imbalance) | |
3859 | { | |
3860 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3861 | unsigned int imbn = 2; | |
3862 | ||
3863 | if (sds->this_nr_running) { | |
3864 | sds->this_load_per_task /= sds->this_nr_running; | |
3865 | if (sds->busiest_load_per_task > | |
3866 | sds->this_load_per_task) | |
3867 | imbn = 1; | |
3868 | } else | |
3869 | sds->this_load_per_task = | |
3870 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3871 | |
2e6f44ae GS |
3872 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3873 | sds->busiest_load_per_task * imbn) { | |
3874 | *imbalance = sds->busiest_load_per_task; | |
3875 | return; | |
3876 | } | |
908a7c1b | 3877 | |
1da177e4 | 3878 | /* |
2e6f44ae GS |
3879 | * OK, we don't have enough imbalance to justify moving tasks, |
3880 | * however we may be able to increase total CPU power used by | |
3881 | * moving them. | |
1da177e4 | 3882 | */ |
2dd73a4f | 3883 | |
2e6f44ae GS |
3884 | pwr_now += sds->busiest->__cpu_power * |
3885 | min(sds->busiest_load_per_task, sds->max_load); | |
3886 | pwr_now += sds->this->__cpu_power * | |
3887 | min(sds->this_load_per_task, sds->this_load); | |
3888 | pwr_now /= SCHED_LOAD_SCALE; | |
3889 | ||
3890 | /* Amount of load we'd subtract */ | |
3891 | tmp = sg_div_cpu_power(sds->busiest, | |
3892 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3893 | if (sds->max_load > tmp) | |
3894 | pwr_move += sds->busiest->__cpu_power * | |
3895 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3896 | ||
3897 | /* Amount of load we'd add */ | |
3898 | if (sds->max_load * sds->busiest->__cpu_power < | |
3899 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3900 | tmp = sg_div_cpu_power(sds->this, | |
3901 | sds->max_load * sds->busiest->__cpu_power); | |
3902 | else | |
3903 | tmp = sg_div_cpu_power(sds->this, | |
3904 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3905 | pwr_move += sds->this->__cpu_power * | |
3906 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3907 | pwr_move /= SCHED_LOAD_SCALE; | |
3908 | ||
3909 | /* Move if we gain throughput */ | |
3910 | if (pwr_move > pwr_now) | |
3911 | *imbalance = sds->busiest_load_per_task; | |
3912 | } | |
dbc523a3 GS |
3913 | |
3914 | /** | |
3915 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3916 | * groups of a given sched_domain during load balance. | |
3917 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3918 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3919 | * @imbalance: The variable to store the imbalance. | |
3920 | */ | |
3921 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3922 | unsigned long *imbalance) | |
3923 | { | |
3924 | unsigned long max_pull; | |
2dd73a4f PW |
3925 | /* |
3926 | * In the presence of smp nice balancing, certain scenarios can have | |
3927 | * max load less than avg load(as we skip the groups at or below | |
3928 | * its cpu_power, while calculating max_load..) | |
3929 | */ | |
dbc523a3 | 3930 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3931 | *imbalance = 0; |
dbc523a3 | 3932 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3933 | } |
0c117f1b SS |
3934 | |
3935 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3936 | max_pull = min(sds->max_load - sds->avg_load, |
3937 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3938 | |
1da177e4 | 3939 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3940 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3941 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3942 | / SCHED_LOAD_SCALE; |
3943 | ||
2dd73a4f PW |
3944 | /* |
3945 | * if *imbalance is less than the average load per runnable task | |
3946 | * there is no gaurantee that any tasks will be moved so we'll have | |
3947 | * a think about bumping its value to force at least one task to be | |
3948 | * moved | |
3949 | */ | |
dbc523a3 GS |
3950 | if (*imbalance < sds->busiest_load_per_task) |
3951 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3952 | |
dbc523a3 | 3953 | } |
37abe198 | 3954 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3955 | |
b7bb4c9b GS |
3956 | /** |
3957 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3958 | * if there is an imbalance. If there isn't an imbalance, and | |
3959 | * the user has opted for power-savings, it returns a group whose | |
3960 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3961 | * such a group exists. | |
3962 | * | |
3963 | * Also calculates the amount of weighted load which should be moved | |
3964 | * to restore balance. | |
3965 | * | |
3966 | * @sd: The sched_domain whose busiest group is to be returned. | |
3967 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3968 | * @imbalance: Variable which stores amount of weighted load which should | |
3969 | * be moved to restore balance/put a group to idle. | |
3970 | * @idle: The idle status of this_cpu. | |
3971 | * @sd_idle: The idleness of sd | |
3972 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3973 | * @balance: Pointer to a variable indicating if this_cpu | |
3974 | * is the appropriate cpu to perform load balancing at this_level. | |
3975 | * | |
3976 | * Returns: - the busiest group if imbalance exists. | |
3977 | * - If no imbalance and user has opted for power-savings balance, | |
3978 | * return the least loaded group whose CPUs can be | |
3979 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3980 | */ |
3981 | static struct sched_group * | |
3982 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3983 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3984 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3985 | { | |
3986 | struct sd_lb_stats sds; | |
1da177e4 | 3987 | |
37abe198 | 3988 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3989 | |
37abe198 GS |
3990 | /* |
3991 | * Compute the various statistics relavent for load balancing at | |
3992 | * this level. | |
3993 | */ | |
3994 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3995 | balance, &sds); | |
3996 | ||
b7bb4c9b GS |
3997 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3998 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3999 | * at this level. | |
4000 | * 2) There is no busy sibling group to pull from. | |
4001 | * 3) This group is the busiest group. | |
4002 | * 4) This group is more busy than the avg busieness at this | |
4003 | * sched_domain. | |
4004 | * 5) The imbalance is within the specified limit. | |
4005 | * 6) Any rebalance would lead to ping-pong | |
4006 | */ | |
37abe198 GS |
4007 | if (balance && !(*balance)) |
4008 | goto ret; | |
1da177e4 | 4009 | |
b7bb4c9b GS |
4010 | if (!sds.busiest || sds.busiest_nr_running == 0) |
4011 | goto out_balanced; | |
1da177e4 | 4012 | |
b7bb4c9b | 4013 | if (sds.this_load >= sds.max_load) |
1da177e4 | 4014 | goto out_balanced; |
1da177e4 | 4015 | |
222d656d | 4016 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 4017 | |
b7bb4c9b GS |
4018 | if (sds.this_load >= sds.avg_load) |
4019 | goto out_balanced; | |
4020 | ||
4021 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
4022 | goto out_balanced; |
4023 | ||
222d656d GS |
4024 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
4025 | if (sds.group_imb) | |
4026 | sds.busiest_load_per_task = | |
4027 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 4028 | |
1da177e4 LT |
4029 | /* |
4030 | * We're trying to get all the cpus to the average_load, so we don't | |
4031 | * want to push ourselves above the average load, nor do we wish to | |
4032 | * reduce the max loaded cpu below the average load, as either of these | |
4033 | * actions would just result in more rebalancing later, and ping-pong | |
4034 | * tasks around. Thus we look for the minimum possible imbalance. | |
4035 | * Negative imbalances (*we* are more loaded than anyone else) will | |
4036 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 4037 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
4038 | * appear as very large values with unsigned longs. |
4039 | */ | |
222d656d | 4040 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
4041 | goto out_balanced; |
4042 | ||
dbc523a3 GS |
4043 | /* Looks like there is an imbalance. Compute it */ |
4044 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 4045 | return sds.busiest; |
1da177e4 LT |
4046 | |
4047 | out_balanced: | |
c071df18 GS |
4048 | /* |
4049 | * There is no obvious imbalance. But check if we can do some balancing | |
4050 | * to save power. | |
4051 | */ | |
4052 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
4053 | return sds.busiest; | |
783609c6 | 4054 | ret: |
1da177e4 LT |
4055 | *imbalance = 0; |
4056 | return NULL; | |
4057 | } | |
4058 | ||
4059 | /* | |
4060 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4061 | */ | |
70b97a7f | 4062 | static struct rq * |
d15bcfdb | 4063 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4064 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4065 | { |
70b97a7f | 4066 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4067 | unsigned long max_load = 0; |
1da177e4 LT |
4068 | int i; |
4069 | ||
758b2cdc | 4070 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 4071 | unsigned long wl; |
0a2966b4 | 4072 | |
96f874e2 | 4073 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4074 | continue; |
4075 | ||
48f24c4d | 4076 | rq = cpu_rq(i); |
dd41f596 | 4077 | wl = weighted_cpuload(i); |
2dd73a4f | 4078 | |
dd41f596 | 4079 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4080 | continue; |
1da177e4 | 4081 | |
dd41f596 IM |
4082 | if (wl > max_load) { |
4083 | max_load = wl; | |
48f24c4d | 4084 | busiest = rq; |
1da177e4 LT |
4085 | } |
4086 | } | |
4087 | ||
4088 | return busiest; | |
4089 | } | |
4090 | ||
77391d71 NP |
4091 | /* |
4092 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4093 | * so long as it is large enough. | |
4094 | */ | |
4095 | #define MAX_PINNED_INTERVAL 512 | |
4096 | ||
df7c8e84 RR |
4097 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4098 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4099 | ||
1da177e4 LT |
4100 | /* |
4101 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4102 | * tasks if there is an imbalance. | |
1da177e4 | 4103 | */ |
70b97a7f | 4104 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4105 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4106 | int *balance) |
1da177e4 | 4107 | { |
43010659 | 4108 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4109 | struct sched_group *group; |
1da177e4 | 4110 | unsigned long imbalance; |
70b97a7f | 4111 | struct rq *busiest; |
fe2eea3f | 4112 | unsigned long flags; |
df7c8e84 | 4113 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4114 | |
96f874e2 | 4115 | cpumask_setall(cpus); |
7c16ec58 | 4116 | |
89c4710e SS |
4117 | /* |
4118 | * When power savings policy is enabled for the parent domain, idle | |
4119 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4120 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4121 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4122 | */ |
d15bcfdb | 4123 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4124 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4125 | sd_idle = 1; |
1da177e4 | 4126 | |
2d72376b | 4127 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4128 | |
0a2966b4 | 4129 | redo: |
c8cba857 | 4130 | update_shares(sd); |
0a2966b4 | 4131 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4132 | cpus, balance); |
783609c6 | 4133 | |
06066714 | 4134 | if (*balance == 0) |
783609c6 | 4135 | goto out_balanced; |
783609c6 | 4136 | |
1da177e4 LT |
4137 | if (!group) { |
4138 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4139 | goto out_balanced; | |
4140 | } | |
4141 | ||
7c16ec58 | 4142 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4143 | if (!busiest) { |
4144 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4145 | goto out_balanced; | |
4146 | } | |
4147 | ||
db935dbd | 4148 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4149 | |
4150 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4151 | ||
43010659 | 4152 | ld_moved = 0; |
1da177e4 LT |
4153 | if (busiest->nr_running > 1) { |
4154 | /* | |
4155 | * Attempt to move tasks. If find_busiest_group has found | |
4156 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4157 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4158 | * correctly treated as an imbalance. |
4159 | */ | |
fe2eea3f | 4160 | local_irq_save(flags); |
e17224bf | 4161 | double_rq_lock(this_rq, busiest); |
43010659 | 4162 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4163 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4164 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4165 | local_irq_restore(flags); |
81026794 | 4166 | |
46cb4b7c SS |
4167 | /* |
4168 | * some other cpu did the load balance for us. | |
4169 | */ | |
43010659 | 4170 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4171 | resched_cpu(this_cpu); |
4172 | ||
81026794 | 4173 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4174 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4175 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4176 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4177 | goto redo; |
81026794 | 4178 | goto out_balanced; |
0a2966b4 | 4179 | } |
1da177e4 | 4180 | } |
81026794 | 4181 | |
43010659 | 4182 | if (!ld_moved) { |
1da177e4 LT |
4183 | schedstat_inc(sd, lb_failed[idle]); |
4184 | sd->nr_balance_failed++; | |
4185 | ||
4186 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4187 | |
fe2eea3f | 4188 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4189 | |
4190 | /* don't kick the migration_thread, if the curr | |
4191 | * task on busiest cpu can't be moved to this_cpu | |
4192 | */ | |
96f874e2 RR |
4193 | if (!cpumask_test_cpu(this_cpu, |
4194 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4195 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4196 | all_pinned = 1; |
4197 | goto out_one_pinned; | |
4198 | } | |
4199 | ||
1da177e4 LT |
4200 | if (!busiest->active_balance) { |
4201 | busiest->active_balance = 1; | |
4202 | busiest->push_cpu = this_cpu; | |
81026794 | 4203 | active_balance = 1; |
1da177e4 | 4204 | } |
fe2eea3f | 4205 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4206 | if (active_balance) |
1da177e4 LT |
4207 | wake_up_process(busiest->migration_thread); |
4208 | ||
4209 | /* | |
4210 | * We've kicked active balancing, reset the failure | |
4211 | * counter. | |
4212 | */ | |
39507451 | 4213 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4214 | } |
81026794 | 4215 | } else |
1da177e4 LT |
4216 | sd->nr_balance_failed = 0; |
4217 | ||
81026794 | 4218 | if (likely(!active_balance)) { |
1da177e4 LT |
4219 | /* We were unbalanced, so reset the balancing interval */ |
4220 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4221 | } else { |
4222 | /* | |
4223 | * If we've begun active balancing, start to back off. This | |
4224 | * case may not be covered by the all_pinned logic if there | |
4225 | * is only 1 task on the busy runqueue (because we don't call | |
4226 | * move_tasks). | |
4227 | */ | |
4228 | if (sd->balance_interval < sd->max_interval) | |
4229 | sd->balance_interval *= 2; | |
1da177e4 LT |
4230 | } |
4231 | ||
43010659 | 4232 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4233 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4234 | ld_moved = -1; |
4235 | ||
4236 | goto out; | |
1da177e4 LT |
4237 | |
4238 | out_balanced: | |
1da177e4 LT |
4239 | schedstat_inc(sd, lb_balanced[idle]); |
4240 | ||
16cfb1c0 | 4241 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4242 | |
4243 | out_one_pinned: | |
1da177e4 | 4244 | /* tune up the balancing interval */ |
77391d71 NP |
4245 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4246 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4247 | sd->balance_interval *= 2; |
4248 | ||
48f24c4d | 4249 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4250 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4251 | ld_moved = -1; |
4252 | else | |
4253 | ld_moved = 0; | |
4254 | out: | |
c8cba857 PZ |
4255 | if (ld_moved) |
4256 | update_shares(sd); | |
c09595f6 | 4257 | return ld_moved; |
1da177e4 LT |
4258 | } |
4259 | ||
4260 | /* | |
4261 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4262 | * tasks if there is an imbalance. | |
4263 | * | |
d15bcfdb | 4264 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4265 | * this_rq is locked. |
4266 | */ | |
48f24c4d | 4267 | static int |
df7c8e84 | 4268 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4269 | { |
4270 | struct sched_group *group; | |
70b97a7f | 4271 | struct rq *busiest = NULL; |
1da177e4 | 4272 | unsigned long imbalance; |
43010659 | 4273 | int ld_moved = 0; |
5969fe06 | 4274 | int sd_idle = 0; |
969bb4e4 | 4275 | int all_pinned = 0; |
df7c8e84 | 4276 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4277 | |
96f874e2 | 4278 | cpumask_setall(cpus); |
5969fe06 | 4279 | |
89c4710e SS |
4280 | /* |
4281 | * When power savings policy is enabled for the parent domain, idle | |
4282 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4283 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4284 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4285 | */ |
4286 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4287 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4288 | sd_idle = 1; |
1da177e4 | 4289 | |
2d72376b | 4290 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4291 | redo: |
3e5459b4 | 4292 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4293 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4294 | &sd_idle, cpus, NULL); |
1da177e4 | 4295 | if (!group) { |
d15bcfdb | 4296 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4297 | goto out_balanced; |
1da177e4 LT |
4298 | } |
4299 | ||
7c16ec58 | 4300 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4301 | if (!busiest) { |
d15bcfdb | 4302 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4303 | goto out_balanced; |
1da177e4 LT |
4304 | } |
4305 | ||
db935dbd NP |
4306 | BUG_ON(busiest == this_rq); |
4307 | ||
d15bcfdb | 4308 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4309 | |
43010659 | 4310 | ld_moved = 0; |
d6d5cfaf NP |
4311 | if (busiest->nr_running > 1) { |
4312 | /* Attempt to move tasks */ | |
4313 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4314 | /* this_rq->clock is already updated */ |
4315 | update_rq_clock(busiest); | |
43010659 | 4316 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4317 | imbalance, sd, CPU_NEWLY_IDLE, |
4318 | &all_pinned); | |
1b12bbc7 | 4319 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4320 | |
969bb4e4 | 4321 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4322 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4323 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4324 | goto redo; |
4325 | } | |
d6d5cfaf NP |
4326 | } |
4327 | ||
43010659 | 4328 | if (!ld_moved) { |
36dffab6 | 4329 | int active_balance = 0; |
ad273b32 | 4330 | |
d15bcfdb | 4331 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4332 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4333 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4334 | return -1; |
ad273b32 VS |
4335 | |
4336 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4337 | return -1; | |
4338 | ||
4339 | if (sd->nr_balance_failed++ < 2) | |
4340 | return -1; | |
4341 | ||
4342 | /* | |
4343 | * The only task running in a non-idle cpu can be moved to this | |
4344 | * cpu in an attempt to completely freeup the other CPU | |
4345 | * package. The same method used to move task in load_balance() | |
4346 | * have been extended for load_balance_newidle() to speedup | |
4347 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4348 | * | |
4349 | * The package power saving logic comes from | |
4350 | * find_busiest_group(). If there are no imbalance, then | |
4351 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4352 | * f_b_g() will select a group from which a running task may be | |
4353 | * pulled to this cpu in order to make the other package idle. | |
4354 | * If there is no opportunity to make a package idle and if | |
4355 | * there are no imbalance, then f_b_g() will return NULL and no | |
4356 | * action will be taken in load_balance_newidle(). | |
4357 | * | |
4358 | * Under normal task pull operation due to imbalance, there | |
4359 | * will be more than one task in the source run queue and | |
4360 | * move_tasks() will succeed. ld_moved will be true and this | |
4361 | * active balance code will not be triggered. | |
4362 | */ | |
4363 | ||
4364 | /* Lock busiest in correct order while this_rq is held */ | |
4365 | double_lock_balance(this_rq, busiest); | |
4366 | ||
4367 | /* | |
4368 | * don't kick the migration_thread, if the curr | |
4369 | * task on busiest cpu can't be moved to this_cpu | |
4370 | */ | |
6ca09dfc | 4371 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4372 | double_unlock_balance(this_rq, busiest); |
4373 | all_pinned = 1; | |
4374 | return ld_moved; | |
4375 | } | |
4376 | ||
4377 | if (!busiest->active_balance) { | |
4378 | busiest->active_balance = 1; | |
4379 | busiest->push_cpu = this_cpu; | |
4380 | active_balance = 1; | |
4381 | } | |
4382 | ||
4383 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4384 | /* |
4385 | * Should not call ttwu while holding a rq->lock | |
4386 | */ | |
4387 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4388 | if (active_balance) |
4389 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4390 | spin_lock(&this_rq->lock); |
ad273b32 | 4391 | |
5969fe06 | 4392 | } else |
16cfb1c0 | 4393 | sd->nr_balance_failed = 0; |
1da177e4 | 4394 | |
3e5459b4 | 4395 | update_shares_locked(this_rq, sd); |
43010659 | 4396 | return ld_moved; |
16cfb1c0 NP |
4397 | |
4398 | out_balanced: | |
d15bcfdb | 4399 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4400 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4401 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4402 | return -1; |
16cfb1c0 | 4403 | sd->nr_balance_failed = 0; |
48f24c4d | 4404 | |
16cfb1c0 | 4405 | return 0; |
1da177e4 LT |
4406 | } |
4407 | ||
4408 | /* | |
4409 | * idle_balance is called by schedule() if this_cpu is about to become | |
4410 | * idle. Attempts to pull tasks from other CPUs. | |
4411 | */ | |
70b97a7f | 4412 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4413 | { |
4414 | struct sched_domain *sd; | |
efbe027e | 4415 | int pulled_task = 0; |
dd41f596 | 4416 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4417 | |
4418 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4419 | unsigned long interval; |
4420 | ||
4421 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4422 | continue; | |
4423 | ||
4424 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4425 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4426 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4427 | sd); |
92c4ca5c CL |
4428 | |
4429 | interval = msecs_to_jiffies(sd->balance_interval); | |
4430 | if (time_after(next_balance, sd->last_balance + interval)) | |
4431 | next_balance = sd->last_balance + interval; | |
4432 | if (pulled_task) | |
4433 | break; | |
1da177e4 | 4434 | } |
dd41f596 | 4435 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4436 | /* |
4437 | * We are going idle. next_balance may be set based on | |
4438 | * a busy processor. So reset next_balance. | |
4439 | */ | |
4440 | this_rq->next_balance = next_balance; | |
dd41f596 | 4441 | } |
1da177e4 LT |
4442 | } |
4443 | ||
4444 | /* | |
4445 | * active_load_balance is run by migration threads. It pushes running tasks | |
4446 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4447 | * running on each physical CPU where possible, and avoids physical / | |
4448 | * logical imbalances. | |
4449 | * | |
4450 | * Called with busiest_rq locked. | |
4451 | */ | |
70b97a7f | 4452 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4453 | { |
39507451 | 4454 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4455 | struct sched_domain *sd; |
4456 | struct rq *target_rq; | |
39507451 | 4457 | |
48f24c4d | 4458 | /* Is there any task to move? */ |
39507451 | 4459 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4460 | return; |
4461 | ||
4462 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4463 | |
4464 | /* | |
39507451 | 4465 | * This condition is "impossible", if it occurs |
41a2d6cf | 4466 | * we need to fix it. Originally reported by |
39507451 | 4467 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4468 | */ |
39507451 | 4469 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4470 | |
39507451 NP |
4471 | /* move a task from busiest_rq to target_rq */ |
4472 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4473 | update_rq_clock(busiest_rq); |
4474 | update_rq_clock(target_rq); | |
39507451 NP |
4475 | |
4476 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4477 | for_each_domain(target_cpu, sd) { |
39507451 | 4478 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4479 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4480 | break; |
c96d145e | 4481 | } |
39507451 | 4482 | |
48f24c4d | 4483 | if (likely(sd)) { |
2d72376b | 4484 | schedstat_inc(sd, alb_count); |
39507451 | 4485 | |
43010659 PW |
4486 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4487 | sd, CPU_IDLE)) | |
48f24c4d IM |
4488 | schedstat_inc(sd, alb_pushed); |
4489 | else | |
4490 | schedstat_inc(sd, alb_failed); | |
4491 | } | |
1b12bbc7 | 4492 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4493 | } |
4494 | ||
46cb4b7c SS |
4495 | #ifdef CONFIG_NO_HZ |
4496 | static struct { | |
4497 | atomic_t load_balancer; | |
7d1e6a9b | 4498 | cpumask_var_t cpu_mask; |
f711f609 | 4499 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4500 | } nohz ____cacheline_aligned = { |
4501 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4502 | }; |
4503 | ||
eea08f32 AB |
4504 | int get_nohz_load_balancer(void) |
4505 | { | |
4506 | return atomic_read(&nohz.load_balancer); | |
4507 | } | |
4508 | ||
f711f609 GS |
4509 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4510 | /** | |
4511 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4512 | * @cpu: The cpu whose lowest level of sched domain is to | |
4513 | * be returned. | |
4514 | * @flag: The flag to check for the lowest sched_domain | |
4515 | * for the given cpu. | |
4516 | * | |
4517 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4518 | */ | |
4519 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4520 | { | |
4521 | struct sched_domain *sd; | |
4522 | ||
4523 | for_each_domain(cpu, sd) | |
4524 | if (sd && (sd->flags & flag)) | |
4525 | break; | |
4526 | ||
4527 | return sd; | |
4528 | } | |
4529 | ||
4530 | /** | |
4531 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4532 | * @cpu: The cpu whose domains we're iterating over. | |
4533 | * @sd: variable holding the value of the power_savings_sd | |
4534 | * for cpu. | |
4535 | * @flag: The flag to filter the sched_domains to be iterated. | |
4536 | * | |
4537 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4538 | * set, starting from the lowest sched_domain to the highest. | |
4539 | */ | |
4540 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4541 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4542 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4543 | ||
4544 | /** | |
4545 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4546 | * @ilb_group: group to be checked for semi-idleness | |
4547 | * | |
4548 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4549 | * | |
4550 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4551 | * and atleast one non-idle CPU. This helper function checks if the given | |
4552 | * sched_group is semi-idle or not. | |
4553 | */ | |
4554 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4555 | { | |
4556 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4557 | sched_group_cpus(ilb_group)); | |
4558 | ||
4559 | /* | |
4560 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4561 | * and atleast one idle cpu. | |
4562 | */ | |
4563 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4564 | return 0; | |
4565 | ||
4566 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4567 | return 0; | |
4568 | ||
4569 | return 1; | |
4570 | } | |
4571 | /** | |
4572 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4573 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4574 | * | |
4575 | * Returns: Returns the id of the idle load balancer if it exists, | |
4576 | * Else, returns >= nr_cpu_ids. | |
4577 | * | |
4578 | * This algorithm picks the idle load balancer such that it belongs to a | |
4579 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4580 | * completely idle packages/cores just for the purpose of idle load balancing | |
4581 | * when there are other idle cpu's which are better suited for that job. | |
4582 | */ | |
4583 | static int find_new_ilb(int cpu) | |
4584 | { | |
4585 | struct sched_domain *sd; | |
4586 | struct sched_group *ilb_group; | |
4587 | ||
4588 | /* | |
4589 | * Have idle load balancer selection from semi-idle packages only | |
4590 | * when power-aware load balancing is enabled | |
4591 | */ | |
4592 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4593 | goto out_done; | |
4594 | ||
4595 | /* | |
4596 | * Optimize for the case when we have no idle CPUs or only one | |
4597 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4598 | */ | |
4599 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4600 | goto out_done; | |
4601 | ||
4602 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4603 | ilb_group = sd->groups; | |
4604 | ||
4605 | do { | |
4606 | if (is_semi_idle_group(ilb_group)) | |
4607 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4608 | ||
4609 | ilb_group = ilb_group->next; | |
4610 | ||
4611 | } while (ilb_group != sd->groups); | |
4612 | } | |
4613 | ||
4614 | out_done: | |
4615 | return cpumask_first(nohz.cpu_mask); | |
4616 | } | |
4617 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4618 | static inline int find_new_ilb(int call_cpu) | |
4619 | { | |
6e29ec57 | 4620 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4621 | } |
4622 | #endif | |
4623 | ||
7835b98b | 4624 | /* |
46cb4b7c SS |
4625 | * This routine will try to nominate the ilb (idle load balancing) |
4626 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4627 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4628 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4629 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4630 | * arrives... | |
4631 | * | |
4632 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4633 | * for idle load balancing. ilb owner will still be part of | |
4634 | * nohz.cpu_mask.. | |
7835b98b | 4635 | * |
46cb4b7c SS |
4636 | * While stopping the tick, this cpu will become the ilb owner if there |
4637 | * is no other owner. And will be the owner till that cpu becomes busy | |
4638 | * or if all cpus in the system stop their ticks at which point | |
4639 | * there is no need for ilb owner. | |
4640 | * | |
4641 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4642 | * next busy scheduler_tick() | |
4643 | */ | |
4644 | int select_nohz_load_balancer(int stop_tick) | |
4645 | { | |
4646 | int cpu = smp_processor_id(); | |
4647 | ||
4648 | if (stop_tick) { | |
46cb4b7c SS |
4649 | cpu_rq(cpu)->in_nohz_recently = 1; |
4650 | ||
483b4ee6 SS |
4651 | if (!cpu_active(cpu)) { |
4652 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4653 | return 0; | |
4654 | ||
4655 | /* | |
4656 | * If we are going offline and still the leader, | |
4657 | * give up! | |
4658 | */ | |
46cb4b7c SS |
4659 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4660 | BUG(); | |
483b4ee6 | 4661 | |
46cb4b7c SS |
4662 | return 0; |
4663 | } | |
4664 | ||
483b4ee6 SS |
4665 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4666 | ||
46cb4b7c | 4667 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4668 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4669 | if (atomic_read(&nohz.load_balancer) == cpu) |
4670 | atomic_set(&nohz.load_balancer, -1); | |
4671 | return 0; | |
4672 | } | |
4673 | ||
4674 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4675 | /* make me the ilb owner */ | |
4676 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4677 | return 1; | |
e790fb0b GS |
4678 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4679 | int new_ilb; | |
4680 | ||
4681 | if (!(sched_smt_power_savings || | |
4682 | sched_mc_power_savings)) | |
4683 | return 1; | |
4684 | /* | |
4685 | * Check to see if there is a more power-efficient | |
4686 | * ilb. | |
4687 | */ | |
4688 | new_ilb = find_new_ilb(cpu); | |
4689 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4690 | atomic_set(&nohz.load_balancer, -1); | |
4691 | resched_cpu(new_ilb); | |
4692 | return 0; | |
4693 | } | |
46cb4b7c | 4694 | return 1; |
e790fb0b | 4695 | } |
46cb4b7c | 4696 | } else { |
7d1e6a9b | 4697 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4698 | return 0; |
4699 | ||
7d1e6a9b | 4700 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4701 | |
4702 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4703 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4704 | BUG(); | |
4705 | } | |
4706 | return 0; | |
4707 | } | |
4708 | #endif | |
4709 | ||
4710 | static DEFINE_SPINLOCK(balancing); | |
4711 | ||
4712 | /* | |
7835b98b CL |
4713 | * It checks each scheduling domain to see if it is due to be balanced, |
4714 | * and initiates a balancing operation if so. | |
4715 | * | |
4716 | * Balancing parameters are set up in arch_init_sched_domains. | |
4717 | */ | |
a9957449 | 4718 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4719 | { |
46cb4b7c SS |
4720 | int balance = 1; |
4721 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4722 | unsigned long interval; |
4723 | struct sched_domain *sd; | |
46cb4b7c | 4724 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4725 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4726 | int update_next_balance = 0; |
d07355f5 | 4727 | int need_serialize; |
1da177e4 | 4728 | |
46cb4b7c | 4729 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4730 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4731 | continue; | |
4732 | ||
4733 | interval = sd->balance_interval; | |
d15bcfdb | 4734 | if (idle != CPU_IDLE) |
1da177e4 LT |
4735 | interval *= sd->busy_factor; |
4736 | ||
4737 | /* scale ms to jiffies */ | |
4738 | interval = msecs_to_jiffies(interval); | |
4739 | if (unlikely(!interval)) | |
4740 | interval = 1; | |
dd41f596 IM |
4741 | if (interval > HZ*NR_CPUS/10) |
4742 | interval = HZ*NR_CPUS/10; | |
4743 | ||
d07355f5 | 4744 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4745 | |
d07355f5 | 4746 | if (need_serialize) { |
08c183f3 CL |
4747 | if (!spin_trylock(&balancing)) |
4748 | goto out; | |
4749 | } | |
4750 | ||
c9819f45 | 4751 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4752 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4753 | /* |
4754 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4755 | * longer idle, or one of our SMT siblings is |
4756 | * not idle. | |
4757 | */ | |
d15bcfdb | 4758 | idle = CPU_NOT_IDLE; |
1da177e4 | 4759 | } |
1bd77f2d | 4760 | sd->last_balance = jiffies; |
1da177e4 | 4761 | } |
d07355f5 | 4762 | if (need_serialize) |
08c183f3 CL |
4763 | spin_unlock(&balancing); |
4764 | out: | |
f549da84 | 4765 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4766 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4767 | update_next_balance = 1; |
4768 | } | |
783609c6 SS |
4769 | |
4770 | /* | |
4771 | * Stop the load balance at this level. There is another | |
4772 | * CPU in our sched group which is doing load balancing more | |
4773 | * actively. | |
4774 | */ | |
4775 | if (!balance) | |
4776 | break; | |
1da177e4 | 4777 | } |
f549da84 SS |
4778 | |
4779 | /* | |
4780 | * next_balance will be updated only when there is a need. | |
4781 | * When the cpu is attached to null domain for ex, it will not be | |
4782 | * updated. | |
4783 | */ | |
4784 | if (likely(update_next_balance)) | |
4785 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4786 | } |
4787 | ||
4788 | /* | |
4789 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4790 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4791 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4792 | */ | |
4793 | static void run_rebalance_domains(struct softirq_action *h) | |
4794 | { | |
dd41f596 IM |
4795 | int this_cpu = smp_processor_id(); |
4796 | struct rq *this_rq = cpu_rq(this_cpu); | |
4797 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4798 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4799 | |
dd41f596 | 4800 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4801 | |
4802 | #ifdef CONFIG_NO_HZ | |
4803 | /* | |
4804 | * If this cpu is the owner for idle load balancing, then do the | |
4805 | * balancing on behalf of the other idle cpus whose ticks are | |
4806 | * stopped. | |
4807 | */ | |
dd41f596 IM |
4808 | if (this_rq->idle_at_tick && |
4809 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4810 | struct rq *rq; |
4811 | int balance_cpu; | |
4812 | ||
7d1e6a9b RR |
4813 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4814 | if (balance_cpu == this_cpu) | |
4815 | continue; | |
4816 | ||
46cb4b7c SS |
4817 | /* |
4818 | * If this cpu gets work to do, stop the load balancing | |
4819 | * work being done for other cpus. Next load | |
4820 | * balancing owner will pick it up. | |
4821 | */ | |
4822 | if (need_resched()) | |
4823 | break; | |
4824 | ||
de0cf899 | 4825 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4826 | |
4827 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4828 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4829 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4830 | } |
4831 | } | |
4832 | #endif | |
4833 | } | |
4834 | ||
8a0be9ef FW |
4835 | static inline int on_null_domain(int cpu) |
4836 | { | |
4837 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4838 | } | |
4839 | ||
46cb4b7c SS |
4840 | /* |
4841 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4842 | * | |
4843 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4844 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4845 | * if the whole system is idle. | |
4846 | */ | |
dd41f596 | 4847 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4848 | { |
46cb4b7c SS |
4849 | #ifdef CONFIG_NO_HZ |
4850 | /* | |
4851 | * If we were in the nohz mode recently and busy at the current | |
4852 | * scheduler tick, then check if we need to nominate new idle | |
4853 | * load balancer. | |
4854 | */ | |
4855 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4856 | rq->in_nohz_recently = 0; | |
4857 | ||
4858 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4859 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4860 | atomic_set(&nohz.load_balancer, -1); |
4861 | } | |
4862 | ||
4863 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4864 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4865 | |
434d53b0 | 4866 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4867 | resched_cpu(ilb); |
4868 | } | |
4869 | } | |
4870 | ||
4871 | /* | |
4872 | * If this cpu is idle and doing idle load balancing for all the | |
4873 | * cpus with ticks stopped, is it time for that to stop? | |
4874 | */ | |
4875 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4876 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4877 | resched_cpu(cpu); |
4878 | return; | |
4879 | } | |
4880 | ||
4881 | /* | |
4882 | * If this cpu is idle and the idle load balancing is done by | |
4883 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4884 | */ | |
4885 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4886 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4887 | return; |
4888 | #endif | |
8a0be9ef FW |
4889 | /* Don't need to rebalance while attached to NULL domain */ |
4890 | if (time_after_eq(jiffies, rq->next_balance) && | |
4891 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4892 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4893 | } |
dd41f596 IM |
4894 | |
4895 | #else /* CONFIG_SMP */ | |
4896 | ||
1da177e4 LT |
4897 | /* |
4898 | * on UP we do not need to balance between CPUs: | |
4899 | */ | |
70b97a7f | 4900 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4901 | { |
4902 | } | |
dd41f596 | 4903 | |
1da177e4 LT |
4904 | #endif |
4905 | ||
1da177e4 LT |
4906 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4907 | ||
4908 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4909 | ||
4910 | /* | |
c5f8d995 | 4911 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4912 | * @p in case that task is currently running. |
c5f8d995 HS |
4913 | * |
4914 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4915 | */ |
c5f8d995 HS |
4916 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4917 | { | |
4918 | u64 ns = 0; | |
4919 | ||
4920 | if (task_current(rq, p)) { | |
4921 | update_rq_clock(rq); | |
4922 | ns = rq->clock - p->se.exec_start; | |
4923 | if ((s64)ns < 0) | |
4924 | ns = 0; | |
4925 | } | |
4926 | ||
4927 | return ns; | |
4928 | } | |
4929 | ||
bb34d92f | 4930 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4931 | { |
1da177e4 | 4932 | unsigned long flags; |
41b86e9c | 4933 | struct rq *rq; |
bb34d92f | 4934 | u64 ns = 0; |
48f24c4d | 4935 | |
41b86e9c | 4936 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4937 | ns = do_task_delta_exec(p, rq); |
4938 | task_rq_unlock(rq, &flags); | |
1508487e | 4939 | |
c5f8d995 HS |
4940 | return ns; |
4941 | } | |
f06febc9 | 4942 | |
c5f8d995 HS |
4943 | /* |
4944 | * Return accounted runtime for the task. | |
4945 | * In case the task is currently running, return the runtime plus current's | |
4946 | * pending runtime that have not been accounted yet. | |
4947 | */ | |
4948 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4949 | { | |
4950 | unsigned long flags; | |
4951 | struct rq *rq; | |
4952 | u64 ns = 0; | |
4953 | ||
4954 | rq = task_rq_lock(p, &flags); | |
4955 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4956 | task_rq_unlock(rq, &flags); | |
4957 | ||
4958 | return ns; | |
4959 | } | |
48f24c4d | 4960 | |
c5f8d995 HS |
4961 | /* |
4962 | * Return sum_exec_runtime for the thread group. | |
4963 | * In case the task is currently running, return the sum plus current's | |
4964 | * pending runtime that have not been accounted yet. | |
4965 | * | |
4966 | * Note that the thread group might have other running tasks as well, | |
4967 | * so the return value not includes other pending runtime that other | |
4968 | * running tasks might have. | |
4969 | */ | |
4970 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4971 | { | |
4972 | struct task_cputime totals; | |
4973 | unsigned long flags; | |
4974 | struct rq *rq; | |
4975 | u64 ns; | |
4976 | ||
4977 | rq = task_rq_lock(p, &flags); | |
4978 | thread_group_cputime(p, &totals); | |
4979 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4980 | task_rq_unlock(rq, &flags); |
48f24c4d | 4981 | |
1da177e4 LT |
4982 | return ns; |
4983 | } | |
4984 | ||
1da177e4 LT |
4985 | /* |
4986 | * Account user cpu time to a process. | |
4987 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4988 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4989 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4990 | */ |
457533a7 MS |
4991 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4992 | cputime_t cputime_scaled) | |
1da177e4 LT |
4993 | { |
4994 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4995 | cputime64_t tmp; | |
4996 | ||
457533a7 | 4997 | /* Add user time to process. */ |
1da177e4 | 4998 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4999 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5000 | account_group_user_time(p, cputime); |
1da177e4 LT |
5001 | |
5002 | /* Add user time to cpustat. */ | |
5003 | tmp = cputime_to_cputime64(cputime); | |
5004 | if (TASK_NICE(p) > 0) | |
5005 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5006 | else | |
5007 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
5008 | |
5009 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
5010 | /* Account for user time used */ |
5011 | acct_update_integrals(p); | |
1da177e4 LT |
5012 | } |
5013 | ||
94886b84 LV |
5014 | /* |
5015 | * Account guest cpu time to a process. | |
5016 | * @p: the process that the cpu time gets accounted to | |
5017 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 5018 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 5019 | */ |
457533a7 MS |
5020 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
5021 | cputime_t cputime_scaled) | |
94886b84 LV |
5022 | { |
5023 | cputime64_t tmp; | |
5024 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5025 | ||
5026 | tmp = cputime_to_cputime64(cputime); | |
5027 | ||
457533a7 | 5028 | /* Add guest time to process. */ |
94886b84 | 5029 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5030 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5031 | account_group_user_time(p, cputime); |
94886b84 LV |
5032 | p->gtime = cputime_add(p->gtime, cputime); |
5033 | ||
457533a7 | 5034 | /* Add guest time to cpustat. */ |
94886b84 LV |
5035 | cpustat->user = cputime64_add(cpustat->user, tmp); |
5036 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
5037 | } | |
5038 | ||
1da177e4 LT |
5039 | /* |
5040 | * Account system cpu time to a process. | |
5041 | * @p: the process that the cpu time gets accounted to | |
5042 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
5043 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 5044 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
5045 | */ |
5046 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 5047 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
5048 | { |
5049 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
5050 | cputime64_t tmp; |
5051 | ||
983ed7a6 | 5052 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 5053 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
5054 | return; |
5055 | } | |
94886b84 | 5056 | |
457533a7 | 5057 | /* Add system time to process. */ |
1da177e4 | 5058 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5059 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5060 | account_group_system_time(p, cputime); |
1da177e4 LT |
5061 | |
5062 | /* Add system time to cpustat. */ | |
5063 | tmp = cputime_to_cputime64(cputime); | |
5064 | if (hardirq_count() - hardirq_offset) | |
5065 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5066 | else if (softirq_count()) | |
5067 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5068 | else |
79741dd3 MS |
5069 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5070 | ||
ef12fefa BR |
5071 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5072 | ||
1da177e4 LT |
5073 | /* Account for system time used */ |
5074 | acct_update_integrals(p); | |
1da177e4 LT |
5075 | } |
5076 | ||
c66f08be | 5077 | /* |
1da177e4 | 5078 | * Account for involuntary wait time. |
1da177e4 | 5079 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5080 | */ |
79741dd3 | 5081 | void account_steal_time(cputime_t cputime) |
c66f08be | 5082 | { |
79741dd3 MS |
5083 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5084 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5085 | ||
5086 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5087 | } |
5088 | ||
1da177e4 | 5089 | /* |
79741dd3 MS |
5090 | * Account for idle time. |
5091 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5092 | */ |
79741dd3 | 5093 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5094 | { |
5095 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5096 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5097 | struct rq *rq = this_rq(); |
1da177e4 | 5098 | |
79741dd3 MS |
5099 | if (atomic_read(&rq->nr_iowait) > 0) |
5100 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5101 | else | |
5102 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5103 | } |
5104 | ||
79741dd3 MS |
5105 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5106 | ||
5107 | /* | |
5108 | * Account a single tick of cpu time. | |
5109 | * @p: the process that the cpu time gets accounted to | |
5110 | * @user_tick: indicates if the tick is a user or a system tick | |
5111 | */ | |
5112 | void account_process_tick(struct task_struct *p, int user_tick) | |
5113 | { | |
5114 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5115 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5116 | struct rq *rq = this_rq(); | |
5117 | ||
5118 | if (user_tick) | |
5119 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5120 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5121 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5122 | one_jiffy_scaled); | |
5123 | else | |
5124 | account_idle_time(one_jiffy); | |
5125 | } | |
5126 | ||
5127 | /* | |
5128 | * Account multiple ticks of steal time. | |
5129 | * @p: the process from which the cpu time has been stolen | |
5130 | * @ticks: number of stolen ticks | |
5131 | */ | |
5132 | void account_steal_ticks(unsigned long ticks) | |
5133 | { | |
5134 | account_steal_time(jiffies_to_cputime(ticks)); | |
5135 | } | |
5136 | ||
5137 | /* | |
5138 | * Account multiple ticks of idle time. | |
5139 | * @ticks: number of stolen ticks | |
5140 | */ | |
5141 | void account_idle_ticks(unsigned long ticks) | |
5142 | { | |
5143 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5144 | } |
5145 | ||
79741dd3 MS |
5146 | #endif |
5147 | ||
49048622 BS |
5148 | /* |
5149 | * Use precise platform statistics if available: | |
5150 | */ | |
5151 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5152 | cputime_t task_utime(struct task_struct *p) | |
5153 | { | |
5154 | return p->utime; | |
5155 | } | |
5156 | ||
5157 | cputime_t task_stime(struct task_struct *p) | |
5158 | { | |
5159 | return p->stime; | |
5160 | } | |
5161 | #else | |
5162 | cputime_t task_utime(struct task_struct *p) | |
5163 | { | |
5164 | clock_t utime = cputime_to_clock_t(p->utime), | |
5165 | total = utime + cputime_to_clock_t(p->stime); | |
5166 | u64 temp; | |
5167 | ||
5168 | /* | |
5169 | * Use CFS's precise accounting: | |
5170 | */ | |
5171 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5172 | ||
5173 | if (total) { | |
5174 | temp *= utime; | |
5175 | do_div(temp, total); | |
5176 | } | |
5177 | utime = (clock_t)temp; | |
5178 | ||
5179 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5180 | return p->prev_utime; | |
5181 | } | |
5182 | ||
5183 | cputime_t task_stime(struct task_struct *p) | |
5184 | { | |
5185 | clock_t stime; | |
5186 | ||
5187 | /* | |
5188 | * Use CFS's precise accounting. (we subtract utime from | |
5189 | * the total, to make sure the total observed by userspace | |
5190 | * grows monotonically - apps rely on that): | |
5191 | */ | |
5192 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5193 | cputime_to_clock_t(task_utime(p)); | |
5194 | ||
5195 | if (stime >= 0) | |
5196 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5197 | ||
5198 | return p->prev_stime; | |
5199 | } | |
5200 | #endif | |
5201 | ||
5202 | inline cputime_t task_gtime(struct task_struct *p) | |
5203 | { | |
5204 | return p->gtime; | |
5205 | } | |
5206 | ||
7835b98b CL |
5207 | /* |
5208 | * This function gets called by the timer code, with HZ frequency. | |
5209 | * We call it with interrupts disabled. | |
5210 | * | |
5211 | * It also gets called by the fork code, when changing the parent's | |
5212 | * timeslices. | |
5213 | */ | |
5214 | void scheduler_tick(void) | |
5215 | { | |
7835b98b CL |
5216 | int cpu = smp_processor_id(); |
5217 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5218 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5219 | |
5220 | sched_clock_tick(); | |
dd41f596 IM |
5221 | |
5222 | spin_lock(&rq->lock); | |
3e51f33f | 5223 | update_rq_clock(rq); |
f1a438d8 | 5224 | update_cpu_load(rq); |
fa85ae24 | 5225 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5226 | spin_unlock(&rq->lock); |
7835b98b | 5227 | |
e220d2dc PZ |
5228 | perf_counter_task_tick(curr, cpu); |
5229 | ||
e418e1c2 | 5230 | #ifdef CONFIG_SMP |
dd41f596 IM |
5231 | rq->idle_at_tick = idle_cpu(cpu); |
5232 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5233 | #endif |
1da177e4 LT |
5234 | } |
5235 | ||
132380a0 | 5236 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5237 | { |
5238 | if (in_lock_functions(addr)) { | |
5239 | addr = CALLER_ADDR2; | |
5240 | if (in_lock_functions(addr)) | |
5241 | addr = CALLER_ADDR3; | |
5242 | } | |
5243 | return addr; | |
5244 | } | |
1da177e4 | 5245 | |
7e49fcce SR |
5246 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5247 | defined(CONFIG_PREEMPT_TRACER)) | |
5248 | ||
43627582 | 5249 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5250 | { |
6cd8a4bb | 5251 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5252 | /* |
5253 | * Underflow? | |
5254 | */ | |
9a11b49a IM |
5255 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5256 | return; | |
6cd8a4bb | 5257 | #endif |
1da177e4 | 5258 | preempt_count() += val; |
6cd8a4bb | 5259 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5260 | /* |
5261 | * Spinlock count overflowing soon? | |
5262 | */ | |
33859f7f MOS |
5263 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5264 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5265 | #endif |
5266 | if (preempt_count() == val) | |
5267 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5268 | } |
5269 | EXPORT_SYMBOL(add_preempt_count); | |
5270 | ||
43627582 | 5271 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5272 | { |
6cd8a4bb | 5273 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5274 | /* |
5275 | * Underflow? | |
5276 | */ | |
01e3eb82 | 5277 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5278 | return; |
1da177e4 LT |
5279 | /* |
5280 | * Is the spinlock portion underflowing? | |
5281 | */ | |
9a11b49a IM |
5282 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5283 | !(preempt_count() & PREEMPT_MASK))) | |
5284 | return; | |
6cd8a4bb | 5285 | #endif |
9a11b49a | 5286 | |
6cd8a4bb SR |
5287 | if (preempt_count() == val) |
5288 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5289 | preempt_count() -= val; |
5290 | } | |
5291 | EXPORT_SYMBOL(sub_preempt_count); | |
5292 | ||
5293 | #endif | |
5294 | ||
5295 | /* | |
dd41f596 | 5296 | * Print scheduling while atomic bug: |
1da177e4 | 5297 | */ |
dd41f596 | 5298 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5299 | { |
838225b4 SS |
5300 | struct pt_regs *regs = get_irq_regs(); |
5301 | ||
5302 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5303 | prev->comm, prev->pid, preempt_count()); | |
5304 | ||
dd41f596 | 5305 | debug_show_held_locks(prev); |
e21f5b15 | 5306 | print_modules(); |
dd41f596 IM |
5307 | if (irqs_disabled()) |
5308 | print_irqtrace_events(prev); | |
838225b4 SS |
5309 | |
5310 | if (regs) | |
5311 | show_regs(regs); | |
5312 | else | |
5313 | dump_stack(); | |
dd41f596 | 5314 | } |
1da177e4 | 5315 | |
dd41f596 IM |
5316 | /* |
5317 | * Various schedule()-time debugging checks and statistics: | |
5318 | */ | |
5319 | static inline void schedule_debug(struct task_struct *prev) | |
5320 | { | |
1da177e4 | 5321 | /* |
41a2d6cf | 5322 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5323 | * schedule() atomically, we ignore that path for now. |
5324 | * Otherwise, whine if we are scheduling when we should not be. | |
5325 | */ | |
3f33a7ce | 5326 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5327 | __schedule_bug(prev); |
5328 | ||
1da177e4 LT |
5329 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5330 | ||
2d72376b | 5331 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5332 | #ifdef CONFIG_SCHEDSTATS |
5333 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5334 | schedstat_inc(this_rq(), bkl_count); |
5335 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5336 | } |
5337 | #endif | |
dd41f596 IM |
5338 | } |
5339 | ||
df1c99d4 MG |
5340 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5341 | { | |
5342 | if (prev->state == TASK_RUNNING) { | |
5343 | u64 runtime = prev->se.sum_exec_runtime; | |
5344 | ||
5345 | runtime -= prev->se.prev_sum_exec_runtime; | |
5346 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5347 | ||
5348 | /* | |
5349 | * In order to avoid avg_overlap growing stale when we are | |
5350 | * indeed overlapping and hence not getting put to sleep, grow | |
5351 | * the avg_overlap on preemption. | |
5352 | * | |
5353 | * We use the average preemption runtime because that | |
5354 | * correlates to the amount of cache footprint a task can | |
5355 | * build up. | |
5356 | */ | |
5357 | update_avg(&prev->se.avg_overlap, runtime); | |
5358 | } | |
5359 | prev->sched_class->put_prev_task(rq, prev); | |
5360 | } | |
5361 | ||
dd41f596 IM |
5362 | /* |
5363 | * Pick up the highest-prio task: | |
5364 | */ | |
5365 | static inline struct task_struct * | |
b67802ea | 5366 | pick_next_task(struct rq *rq) |
dd41f596 | 5367 | { |
5522d5d5 | 5368 | const struct sched_class *class; |
dd41f596 | 5369 | struct task_struct *p; |
1da177e4 LT |
5370 | |
5371 | /* | |
dd41f596 IM |
5372 | * Optimization: we know that if all tasks are in |
5373 | * the fair class we can call that function directly: | |
1da177e4 | 5374 | */ |
dd41f596 | 5375 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5376 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5377 | if (likely(p)) |
5378 | return p; | |
1da177e4 LT |
5379 | } |
5380 | ||
dd41f596 IM |
5381 | class = sched_class_highest; |
5382 | for ( ; ; ) { | |
fb8d4724 | 5383 | p = class->pick_next_task(rq); |
dd41f596 IM |
5384 | if (p) |
5385 | return p; | |
5386 | /* | |
5387 | * Will never be NULL as the idle class always | |
5388 | * returns a non-NULL p: | |
5389 | */ | |
5390 | class = class->next; | |
5391 | } | |
5392 | } | |
1da177e4 | 5393 | |
dd41f596 IM |
5394 | /* |
5395 | * schedule() is the main scheduler function. | |
5396 | */ | |
ff743345 | 5397 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5398 | { |
5399 | struct task_struct *prev, *next; | |
67ca7bde | 5400 | unsigned long *switch_count; |
dd41f596 | 5401 | struct rq *rq; |
31656519 | 5402 | int cpu; |
dd41f596 | 5403 | |
ff743345 PZ |
5404 | need_resched: |
5405 | preempt_disable(); | |
dd41f596 IM |
5406 | cpu = smp_processor_id(); |
5407 | rq = cpu_rq(cpu); | |
5408 | rcu_qsctr_inc(cpu); | |
5409 | prev = rq->curr; | |
5410 | switch_count = &prev->nivcsw; | |
5411 | ||
5412 | release_kernel_lock(prev); | |
5413 | need_resched_nonpreemptible: | |
5414 | ||
5415 | schedule_debug(prev); | |
1da177e4 | 5416 | |
31656519 | 5417 | if (sched_feat(HRTICK)) |
f333fdc9 | 5418 | hrtick_clear(rq); |
8f4d37ec | 5419 | |
8cd162ce | 5420 | spin_lock_irq(&rq->lock); |
3e51f33f | 5421 | update_rq_clock(rq); |
1e819950 | 5422 | clear_tsk_need_resched(prev); |
1da177e4 | 5423 | |
1da177e4 | 5424 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5425 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5426 | prev->state = TASK_RUNNING; |
16882c1e | 5427 | else |
2e1cb74a | 5428 | deactivate_task(rq, prev, 1); |
dd41f596 | 5429 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5430 | } |
5431 | ||
3f029d3c | 5432 | pre_schedule(rq, prev); |
f65eda4f | 5433 | |
dd41f596 | 5434 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5435 | idle_balance(cpu, rq); |
1da177e4 | 5436 | |
df1c99d4 | 5437 | put_prev_task(rq, prev); |
b67802ea | 5438 | next = pick_next_task(rq); |
1da177e4 | 5439 | |
1da177e4 | 5440 | if (likely(prev != next)) { |
673a90a1 | 5441 | sched_info_switch(prev, next); |
564c2b21 | 5442 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5443 | |
1da177e4 LT |
5444 | rq->nr_switches++; |
5445 | rq->curr = next; | |
5446 | ++*switch_count; | |
5447 | ||
3f029d3c | 5448 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5449 | /* |
5450 | * the context switch might have flipped the stack from under | |
5451 | * us, hence refresh the local variables. | |
5452 | */ | |
5453 | cpu = smp_processor_id(); | |
5454 | rq = cpu_rq(cpu); | |
3f029d3c | 5455 | } else |
1da177e4 | 5456 | spin_unlock_irq(&rq->lock); |
da19ab51 | 5457 | |
3f029d3c | 5458 | post_schedule(rq); |
1da177e4 | 5459 | |
8f4d37ec | 5460 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5461 | goto need_resched_nonpreemptible; |
8f4d37ec | 5462 | |
1da177e4 | 5463 | preempt_enable_no_resched(); |
ff743345 | 5464 | if (need_resched()) |
1da177e4 LT |
5465 | goto need_resched; |
5466 | } | |
1da177e4 LT |
5467 | EXPORT_SYMBOL(schedule); |
5468 | ||
0d66bf6d PZ |
5469 | #ifdef CONFIG_SMP |
5470 | /* | |
5471 | * Look out! "owner" is an entirely speculative pointer | |
5472 | * access and not reliable. | |
5473 | */ | |
5474 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5475 | { | |
5476 | unsigned int cpu; | |
5477 | struct rq *rq; | |
5478 | ||
5479 | if (!sched_feat(OWNER_SPIN)) | |
5480 | return 0; | |
5481 | ||
5482 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5483 | /* | |
5484 | * Need to access the cpu field knowing that | |
5485 | * DEBUG_PAGEALLOC could have unmapped it if | |
5486 | * the mutex owner just released it and exited. | |
5487 | */ | |
5488 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5489 | goto out; | |
5490 | #else | |
5491 | cpu = owner->cpu; | |
5492 | #endif | |
5493 | ||
5494 | /* | |
5495 | * Even if the access succeeded (likely case), | |
5496 | * the cpu field may no longer be valid. | |
5497 | */ | |
5498 | if (cpu >= nr_cpumask_bits) | |
5499 | goto out; | |
5500 | ||
5501 | /* | |
5502 | * We need to validate that we can do a | |
5503 | * get_cpu() and that we have the percpu area. | |
5504 | */ | |
5505 | if (!cpu_online(cpu)) | |
5506 | goto out; | |
5507 | ||
5508 | rq = cpu_rq(cpu); | |
5509 | ||
5510 | for (;;) { | |
5511 | /* | |
5512 | * Owner changed, break to re-assess state. | |
5513 | */ | |
5514 | if (lock->owner != owner) | |
5515 | break; | |
5516 | ||
5517 | /* | |
5518 | * Is that owner really running on that cpu? | |
5519 | */ | |
5520 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5521 | return 0; | |
5522 | ||
5523 | cpu_relax(); | |
5524 | } | |
5525 | out: | |
5526 | return 1; | |
5527 | } | |
5528 | #endif | |
5529 | ||
1da177e4 LT |
5530 | #ifdef CONFIG_PREEMPT |
5531 | /* | |
2ed6e34f | 5532 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5533 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5534 | * occur there and call schedule directly. |
5535 | */ | |
5536 | asmlinkage void __sched preempt_schedule(void) | |
5537 | { | |
5538 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5539 | |
1da177e4 LT |
5540 | /* |
5541 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5542 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5543 | */ |
beed33a8 | 5544 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5545 | return; |
5546 | ||
3a5c359a AK |
5547 | do { |
5548 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5549 | schedule(); |
3a5c359a | 5550 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5551 | |
3a5c359a AK |
5552 | /* |
5553 | * Check again in case we missed a preemption opportunity | |
5554 | * between schedule and now. | |
5555 | */ | |
5556 | barrier(); | |
5ed0cec0 | 5557 | } while (need_resched()); |
1da177e4 | 5558 | } |
1da177e4 LT |
5559 | EXPORT_SYMBOL(preempt_schedule); |
5560 | ||
5561 | /* | |
2ed6e34f | 5562 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5563 | * off of irq context. |
5564 | * Note, that this is called and return with irqs disabled. This will | |
5565 | * protect us against recursive calling from irq. | |
5566 | */ | |
5567 | asmlinkage void __sched preempt_schedule_irq(void) | |
5568 | { | |
5569 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5570 | |
2ed6e34f | 5571 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5572 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5573 | ||
3a5c359a AK |
5574 | do { |
5575 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5576 | local_irq_enable(); |
5577 | schedule(); | |
5578 | local_irq_disable(); | |
3a5c359a | 5579 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5580 | |
3a5c359a AK |
5581 | /* |
5582 | * Check again in case we missed a preemption opportunity | |
5583 | * between schedule and now. | |
5584 | */ | |
5585 | barrier(); | |
5ed0cec0 | 5586 | } while (need_resched()); |
1da177e4 LT |
5587 | } |
5588 | ||
5589 | #endif /* CONFIG_PREEMPT */ | |
5590 | ||
95cdf3b7 IM |
5591 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5592 | void *key) | |
1da177e4 | 5593 | { |
48f24c4d | 5594 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5595 | } |
1da177e4 LT |
5596 | EXPORT_SYMBOL(default_wake_function); |
5597 | ||
5598 | /* | |
41a2d6cf IM |
5599 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5600 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5601 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5602 | * | |
5603 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5604 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5605 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5606 | */ | |
78ddb08f | 5607 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5608 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5609 | { |
2e45874c | 5610 | wait_queue_t *curr, *next; |
1da177e4 | 5611 | |
2e45874c | 5612 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5613 | unsigned flags = curr->flags; |
5614 | ||
1da177e4 | 5615 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5616 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5617 | break; |
5618 | } | |
5619 | } | |
5620 | ||
5621 | /** | |
5622 | * __wake_up - wake up threads blocked on a waitqueue. | |
5623 | * @q: the waitqueue | |
5624 | * @mode: which threads | |
5625 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5626 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5627 | * |
5628 | * It may be assumed that this function implies a write memory barrier before | |
5629 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5630 | */ |
7ad5b3a5 | 5631 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5632 | int nr_exclusive, void *key) |
1da177e4 LT |
5633 | { |
5634 | unsigned long flags; | |
5635 | ||
5636 | spin_lock_irqsave(&q->lock, flags); | |
5637 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5638 | spin_unlock_irqrestore(&q->lock, flags); | |
5639 | } | |
1da177e4 LT |
5640 | EXPORT_SYMBOL(__wake_up); |
5641 | ||
5642 | /* | |
5643 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5644 | */ | |
7ad5b3a5 | 5645 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5646 | { |
5647 | __wake_up_common(q, mode, 1, 0, NULL); | |
5648 | } | |
5649 | ||
4ede816a DL |
5650 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5651 | { | |
5652 | __wake_up_common(q, mode, 1, 0, key); | |
5653 | } | |
5654 | ||
1da177e4 | 5655 | /** |
4ede816a | 5656 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5657 | * @q: the waitqueue |
5658 | * @mode: which threads | |
5659 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5660 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5661 | * |
5662 | * The sync wakeup differs that the waker knows that it will schedule | |
5663 | * away soon, so while the target thread will be woken up, it will not | |
5664 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5665 | * with each other. This can prevent needless bouncing between CPUs. | |
5666 | * | |
5667 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5668 | * |
5669 | * It may be assumed that this function implies a write memory barrier before | |
5670 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5671 | */ |
4ede816a DL |
5672 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5673 | int nr_exclusive, void *key) | |
1da177e4 LT |
5674 | { |
5675 | unsigned long flags; | |
5676 | int sync = 1; | |
5677 | ||
5678 | if (unlikely(!q)) | |
5679 | return; | |
5680 | ||
5681 | if (unlikely(!nr_exclusive)) | |
5682 | sync = 0; | |
5683 | ||
5684 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5685 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5686 | spin_unlock_irqrestore(&q->lock, flags); |
5687 | } | |
4ede816a DL |
5688 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5689 | ||
5690 | /* | |
5691 | * __wake_up_sync - see __wake_up_sync_key() | |
5692 | */ | |
5693 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5694 | { | |
5695 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5696 | } | |
1da177e4 LT |
5697 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5698 | ||
65eb3dc6 KD |
5699 | /** |
5700 | * complete: - signals a single thread waiting on this completion | |
5701 | * @x: holds the state of this particular completion | |
5702 | * | |
5703 | * This will wake up a single thread waiting on this completion. Threads will be | |
5704 | * awakened in the same order in which they were queued. | |
5705 | * | |
5706 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5707 | * |
5708 | * It may be assumed that this function implies a write memory barrier before | |
5709 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5710 | */ |
b15136e9 | 5711 | void complete(struct completion *x) |
1da177e4 LT |
5712 | { |
5713 | unsigned long flags; | |
5714 | ||
5715 | spin_lock_irqsave(&x->wait.lock, flags); | |
5716 | x->done++; | |
d9514f6c | 5717 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5718 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5719 | } | |
5720 | EXPORT_SYMBOL(complete); | |
5721 | ||
65eb3dc6 KD |
5722 | /** |
5723 | * complete_all: - signals all threads waiting on this completion | |
5724 | * @x: holds the state of this particular completion | |
5725 | * | |
5726 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5727 | * |
5728 | * It may be assumed that this function implies a write memory barrier before | |
5729 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5730 | */ |
b15136e9 | 5731 | void complete_all(struct completion *x) |
1da177e4 LT |
5732 | { |
5733 | unsigned long flags; | |
5734 | ||
5735 | spin_lock_irqsave(&x->wait.lock, flags); | |
5736 | x->done += UINT_MAX/2; | |
d9514f6c | 5737 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5738 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5739 | } | |
5740 | EXPORT_SYMBOL(complete_all); | |
5741 | ||
8cbbe86d AK |
5742 | static inline long __sched |
5743 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5744 | { |
1da177e4 LT |
5745 | if (!x->done) { |
5746 | DECLARE_WAITQUEUE(wait, current); | |
5747 | ||
5748 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5749 | __add_wait_queue_tail(&x->wait, &wait); | |
5750 | do { | |
94d3d824 | 5751 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5752 | timeout = -ERESTARTSYS; |
5753 | break; | |
8cbbe86d AK |
5754 | } |
5755 | __set_current_state(state); | |
1da177e4 LT |
5756 | spin_unlock_irq(&x->wait.lock); |
5757 | timeout = schedule_timeout(timeout); | |
5758 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5759 | } while (!x->done && timeout); |
1da177e4 | 5760 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5761 | if (!x->done) |
5762 | return timeout; | |
1da177e4 LT |
5763 | } |
5764 | x->done--; | |
ea71a546 | 5765 | return timeout ?: 1; |
1da177e4 | 5766 | } |
1da177e4 | 5767 | |
8cbbe86d AK |
5768 | static long __sched |
5769 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5770 | { |
1da177e4 LT |
5771 | might_sleep(); |
5772 | ||
5773 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5774 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5775 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5776 | return timeout; |
5777 | } | |
1da177e4 | 5778 | |
65eb3dc6 KD |
5779 | /** |
5780 | * wait_for_completion: - waits for completion of a task | |
5781 | * @x: holds the state of this particular completion | |
5782 | * | |
5783 | * This waits to be signaled for completion of a specific task. It is NOT | |
5784 | * interruptible and there is no timeout. | |
5785 | * | |
5786 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5787 | * and interrupt capability. Also see complete(). | |
5788 | */ | |
b15136e9 | 5789 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5790 | { |
5791 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5792 | } |
8cbbe86d | 5793 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5794 | |
65eb3dc6 KD |
5795 | /** |
5796 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5797 | * @x: holds the state of this particular completion | |
5798 | * @timeout: timeout value in jiffies | |
5799 | * | |
5800 | * This waits for either a completion of a specific task to be signaled or for a | |
5801 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5802 | * interruptible. | |
5803 | */ | |
b15136e9 | 5804 | unsigned long __sched |
8cbbe86d | 5805 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5806 | { |
8cbbe86d | 5807 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5808 | } |
8cbbe86d | 5809 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5810 | |
65eb3dc6 KD |
5811 | /** |
5812 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5813 | * @x: holds the state of this particular completion | |
5814 | * | |
5815 | * This waits for completion of a specific task to be signaled. It is | |
5816 | * interruptible. | |
5817 | */ | |
8cbbe86d | 5818 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5819 | { |
51e97990 AK |
5820 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5821 | if (t == -ERESTARTSYS) | |
5822 | return t; | |
5823 | return 0; | |
0fec171c | 5824 | } |
8cbbe86d | 5825 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5826 | |
65eb3dc6 KD |
5827 | /** |
5828 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5829 | * @x: holds the state of this particular completion | |
5830 | * @timeout: timeout value in jiffies | |
5831 | * | |
5832 | * This waits for either a completion of a specific task to be signaled or for a | |
5833 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5834 | */ | |
b15136e9 | 5835 | unsigned long __sched |
8cbbe86d AK |
5836 | wait_for_completion_interruptible_timeout(struct completion *x, |
5837 | unsigned long timeout) | |
0fec171c | 5838 | { |
8cbbe86d | 5839 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5840 | } |
8cbbe86d | 5841 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5842 | |
65eb3dc6 KD |
5843 | /** |
5844 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5845 | * @x: holds the state of this particular completion | |
5846 | * | |
5847 | * This waits to be signaled for completion of a specific task. It can be | |
5848 | * interrupted by a kill signal. | |
5849 | */ | |
009e577e MW |
5850 | int __sched wait_for_completion_killable(struct completion *x) |
5851 | { | |
5852 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5853 | if (t == -ERESTARTSYS) | |
5854 | return t; | |
5855 | return 0; | |
5856 | } | |
5857 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5858 | ||
be4de352 DC |
5859 | /** |
5860 | * try_wait_for_completion - try to decrement a completion without blocking | |
5861 | * @x: completion structure | |
5862 | * | |
5863 | * Returns: 0 if a decrement cannot be done without blocking | |
5864 | * 1 if a decrement succeeded. | |
5865 | * | |
5866 | * If a completion is being used as a counting completion, | |
5867 | * attempt to decrement the counter without blocking. This | |
5868 | * enables us to avoid waiting if the resource the completion | |
5869 | * is protecting is not available. | |
5870 | */ | |
5871 | bool try_wait_for_completion(struct completion *x) | |
5872 | { | |
5873 | int ret = 1; | |
5874 | ||
5875 | spin_lock_irq(&x->wait.lock); | |
5876 | if (!x->done) | |
5877 | ret = 0; | |
5878 | else | |
5879 | x->done--; | |
5880 | spin_unlock_irq(&x->wait.lock); | |
5881 | return ret; | |
5882 | } | |
5883 | EXPORT_SYMBOL(try_wait_for_completion); | |
5884 | ||
5885 | /** | |
5886 | * completion_done - Test to see if a completion has any waiters | |
5887 | * @x: completion structure | |
5888 | * | |
5889 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5890 | * 1 if there are no waiters. | |
5891 | * | |
5892 | */ | |
5893 | bool completion_done(struct completion *x) | |
5894 | { | |
5895 | int ret = 1; | |
5896 | ||
5897 | spin_lock_irq(&x->wait.lock); | |
5898 | if (!x->done) | |
5899 | ret = 0; | |
5900 | spin_unlock_irq(&x->wait.lock); | |
5901 | return ret; | |
5902 | } | |
5903 | EXPORT_SYMBOL(completion_done); | |
5904 | ||
8cbbe86d AK |
5905 | static long __sched |
5906 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5907 | { |
0fec171c IM |
5908 | unsigned long flags; |
5909 | wait_queue_t wait; | |
5910 | ||
5911 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5912 | |
8cbbe86d | 5913 | __set_current_state(state); |
1da177e4 | 5914 | |
8cbbe86d AK |
5915 | spin_lock_irqsave(&q->lock, flags); |
5916 | __add_wait_queue(q, &wait); | |
5917 | spin_unlock(&q->lock); | |
5918 | timeout = schedule_timeout(timeout); | |
5919 | spin_lock_irq(&q->lock); | |
5920 | __remove_wait_queue(q, &wait); | |
5921 | spin_unlock_irqrestore(&q->lock, flags); | |
5922 | ||
5923 | return timeout; | |
5924 | } | |
5925 | ||
5926 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5927 | { | |
5928 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5929 | } |
1da177e4 LT |
5930 | EXPORT_SYMBOL(interruptible_sleep_on); |
5931 | ||
0fec171c | 5932 | long __sched |
95cdf3b7 | 5933 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5934 | { |
8cbbe86d | 5935 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5936 | } |
1da177e4 LT |
5937 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5938 | ||
0fec171c | 5939 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5940 | { |
8cbbe86d | 5941 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5942 | } |
1da177e4 LT |
5943 | EXPORT_SYMBOL(sleep_on); |
5944 | ||
0fec171c | 5945 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5946 | { |
8cbbe86d | 5947 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5948 | } |
1da177e4 LT |
5949 | EXPORT_SYMBOL(sleep_on_timeout); |
5950 | ||
b29739f9 IM |
5951 | #ifdef CONFIG_RT_MUTEXES |
5952 | ||
5953 | /* | |
5954 | * rt_mutex_setprio - set the current priority of a task | |
5955 | * @p: task | |
5956 | * @prio: prio value (kernel-internal form) | |
5957 | * | |
5958 | * This function changes the 'effective' priority of a task. It does | |
5959 | * not touch ->normal_prio like __setscheduler(). | |
5960 | * | |
5961 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5962 | */ | |
36c8b586 | 5963 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5964 | { |
5965 | unsigned long flags; | |
83b699ed | 5966 | int oldprio, on_rq, running; |
70b97a7f | 5967 | struct rq *rq; |
cb469845 | 5968 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5969 | |
5970 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5971 | ||
5972 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5973 | update_rq_clock(rq); |
b29739f9 | 5974 | |
d5f9f942 | 5975 | oldprio = p->prio; |
dd41f596 | 5976 | on_rq = p->se.on_rq; |
051a1d1a | 5977 | running = task_current(rq, p); |
0e1f3483 | 5978 | if (on_rq) |
69be72c1 | 5979 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5980 | if (running) |
5981 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5982 | |
5983 | if (rt_prio(prio)) | |
5984 | p->sched_class = &rt_sched_class; | |
5985 | else | |
5986 | p->sched_class = &fair_sched_class; | |
5987 | ||
b29739f9 IM |
5988 | p->prio = prio; |
5989 | ||
0e1f3483 HS |
5990 | if (running) |
5991 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5992 | if (on_rq) { |
8159f87e | 5993 | enqueue_task(rq, p, 0); |
cb469845 SR |
5994 | |
5995 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5996 | } |
5997 | task_rq_unlock(rq, &flags); | |
5998 | } | |
5999 | ||
6000 | #endif | |
6001 | ||
36c8b586 | 6002 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6003 | { |
dd41f596 | 6004 | int old_prio, delta, on_rq; |
1da177e4 | 6005 | unsigned long flags; |
70b97a7f | 6006 | struct rq *rq; |
1da177e4 LT |
6007 | |
6008 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
6009 | return; | |
6010 | /* | |
6011 | * We have to be careful, if called from sys_setpriority(), | |
6012 | * the task might be in the middle of scheduling on another CPU. | |
6013 | */ | |
6014 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6015 | update_rq_clock(rq); |
1da177e4 LT |
6016 | /* |
6017 | * The RT priorities are set via sched_setscheduler(), but we still | |
6018 | * allow the 'normal' nice value to be set - but as expected | |
6019 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 6020 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 6021 | */ |
e05606d3 | 6022 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
6023 | p->static_prio = NICE_TO_PRIO(nice); |
6024 | goto out_unlock; | |
6025 | } | |
dd41f596 | 6026 | on_rq = p->se.on_rq; |
c09595f6 | 6027 | if (on_rq) |
69be72c1 | 6028 | dequeue_task(rq, p, 0); |
1da177e4 | 6029 | |
1da177e4 | 6030 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 6031 | set_load_weight(p); |
b29739f9 IM |
6032 | old_prio = p->prio; |
6033 | p->prio = effective_prio(p); | |
6034 | delta = p->prio - old_prio; | |
1da177e4 | 6035 | |
dd41f596 | 6036 | if (on_rq) { |
8159f87e | 6037 | enqueue_task(rq, p, 0); |
1da177e4 | 6038 | /* |
d5f9f942 AM |
6039 | * If the task increased its priority or is running and |
6040 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 6041 | */ |
d5f9f942 | 6042 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
6043 | resched_task(rq->curr); |
6044 | } | |
6045 | out_unlock: | |
6046 | task_rq_unlock(rq, &flags); | |
6047 | } | |
1da177e4 LT |
6048 | EXPORT_SYMBOL(set_user_nice); |
6049 | ||
e43379f1 MM |
6050 | /* |
6051 | * can_nice - check if a task can reduce its nice value | |
6052 | * @p: task | |
6053 | * @nice: nice value | |
6054 | */ | |
36c8b586 | 6055 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6056 | { |
024f4747 MM |
6057 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6058 | int nice_rlim = 20 - nice; | |
48f24c4d | 6059 | |
e43379f1 MM |
6060 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6061 | capable(CAP_SYS_NICE)); | |
6062 | } | |
6063 | ||
1da177e4 LT |
6064 | #ifdef __ARCH_WANT_SYS_NICE |
6065 | ||
6066 | /* | |
6067 | * sys_nice - change the priority of the current process. | |
6068 | * @increment: priority increment | |
6069 | * | |
6070 | * sys_setpriority is a more generic, but much slower function that | |
6071 | * does similar things. | |
6072 | */ | |
5add95d4 | 6073 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6074 | { |
48f24c4d | 6075 | long nice, retval; |
1da177e4 LT |
6076 | |
6077 | /* | |
6078 | * Setpriority might change our priority at the same moment. | |
6079 | * We don't have to worry. Conceptually one call occurs first | |
6080 | * and we have a single winner. | |
6081 | */ | |
e43379f1 MM |
6082 | if (increment < -40) |
6083 | increment = -40; | |
1da177e4 LT |
6084 | if (increment > 40) |
6085 | increment = 40; | |
6086 | ||
2b8f836f | 6087 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6088 | if (nice < -20) |
6089 | nice = -20; | |
6090 | if (nice > 19) | |
6091 | nice = 19; | |
6092 | ||
e43379f1 MM |
6093 | if (increment < 0 && !can_nice(current, nice)) |
6094 | return -EPERM; | |
6095 | ||
1da177e4 LT |
6096 | retval = security_task_setnice(current, nice); |
6097 | if (retval) | |
6098 | return retval; | |
6099 | ||
6100 | set_user_nice(current, nice); | |
6101 | return 0; | |
6102 | } | |
6103 | ||
6104 | #endif | |
6105 | ||
6106 | /** | |
6107 | * task_prio - return the priority value of a given task. | |
6108 | * @p: the task in question. | |
6109 | * | |
6110 | * This is the priority value as seen by users in /proc. | |
6111 | * RT tasks are offset by -200. Normal tasks are centered | |
6112 | * around 0, value goes from -16 to +15. | |
6113 | */ | |
36c8b586 | 6114 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6115 | { |
6116 | return p->prio - MAX_RT_PRIO; | |
6117 | } | |
6118 | ||
6119 | /** | |
6120 | * task_nice - return the nice value of a given task. | |
6121 | * @p: the task in question. | |
6122 | */ | |
36c8b586 | 6123 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6124 | { |
6125 | return TASK_NICE(p); | |
6126 | } | |
150d8bed | 6127 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6128 | |
6129 | /** | |
6130 | * idle_cpu - is a given cpu idle currently? | |
6131 | * @cpu: the processor in question. | |
6132 | */ | |
6133 | int idle_cpu(int cpu) | |
6134 | { | |
6135 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6136 | } | |
6137 | ||
1da177e4 LT |
6138 | /** |
6139 | * idle_task - return the idle task for a given cpu. | |
6140 | * @cpu: the processor in question. | |
6141 | */ | |
36c8b586 | 6142 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6143 | { |
6144 | return cpu_rq(cpu)->idle; | |
6145 | } | |
6146 | ||
6147 | /** | |
6148 | * find_process_by_pid - find a process with a matching PID value. | |
6149 | * @pid: the pid in question. | |
6150 | */ | |
a9957449 | 6151 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6152 | { |
228ebcbe | 6153 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6154 | } |
6155 | ||
6156 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6157 | static void |
6158 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6159 | { |
dd41f596 | 6160 | BUG_ON(p->se.on_rq); |
48f24c4d | 6161 | |
1da177e4 | 6162 | p->policy = policy; |
dd41f596 IM |
6163 | switch (p->policy) { |
6164 | case SCHED_NORMAL: | |
6165 | case SCHED_BATCH: | |
6166 | case SCHED_IDLE: | |
6167 | p->sched_class = &fair_sched_class; | |
6168 | break; | |
6169 | case SCHED_FIFO: | |
6170 | case SCHED_RR: | |
6171 | p->sched_class = &rt_sched_class; | |
6172 | break; | |
6173 | } | |
6174 | ||
1da177e4 | 6175 | p->rt_priority = prio; |
b29739f9 IM |
6176 | p->normal_prio = normal_prio(p); |
6177 | /* we are holding p->pi_lock already */ | |
6178 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6179 | set_load_weight(p); |
1da177e4 LT |
6180 | } |
6181 | ||
c69e8d9c DH |
6182 | /* |
6183 | * check the target process has a UID that matches the current process's | |
6184 | */ | |
6185 | static bool check_same_owner(struct task_struct *p) | |
6186 | { | |
6187 | const struct cred *cred = current_cred(), *pcred; | |
6188 | bool match; | |
6189 | ||
6190 | rcu_read_lock(); | |
6191 | pcred = __task_cred(p); | |
6192 | match = (cred->euid == pcred->euid || | |
6193 | cred->euid == pcred->uid); | |
6194 | rcu_read_unlock(); | |
6195 | return match; | |
6196 | } | |
6197 | ||
961ccddd RR |
6198 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6199 | struct sched_param *param, bool user) | |
1da177e4 | 6200 | { |
83b699ed | 6201 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6202 | unsigned long flags; |
cb469845 | 6203 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6204 | struct rq *rq; |
ca94c442 | 6205 | int reset_on_fork; |
1da177e4 | 6206 | |
66e5393a SR |
6207 | /* may grab non-irq protected spin_locks */ |
6208 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6209 | recheck: |
6210 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6211 | if (policy < 0) { |
6212 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6213 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6214 | } else { |
6215 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6216 | policy &= ~SCHED_RESET_ON_FORK; | |
6217 | ||
6218 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6219 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6220 | policy != SCHED_IDLE) | |
6221 | return -EINVAL; | |
6222 | } | |
6223 | ||
1da177e4 LT |
6224 | /* |
6225 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6226 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6227 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6228 | */ |
6229 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6230 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6231 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6232 | return -EINVAL; |
e05606d3 | 6233 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6234 | return -EINVAL; |
6235 | ||
37e4ab3f OC |
6236 | /* |
6237 | * Allow unprivileged RT tasks to decrease priority: | |
6238 | */ | |
961ccddd | 6239 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6240 | if (rt_policy(policy)) { |
8dc3e909 | 6241 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6242 | |
6243 | if (!lock_task_sighand(p, &flags)) | |
6244 | return -ESRCH; | |
6245 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6246 | unlock_task_sighand(p, &flags); | |
6247 | ||
6248 | /* can't set/change the rt policy */ | |
6249 | if (policy != p->policy && !rlim_rtprio) | |
6250 | return -EPERM; | |
6251 | ||
6252 | /* can't increase priority */ | |
6253 | if (param->sched_priority > p->rt_priority && | |
6254 | param->sched_priority > rlim_rtprio) | |
6255 | return -EPERM; | |
6256 | } | |
dd41f596 IM |
6257 | /* |
6258 | * Like positive nice levels, dont allow tasks to | |
6259 | * move out of SCHED_IDLE either: | |
6260 | */ | |
6261 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6262 | return -EPERM; | |
5fe1d75f | 6263 | |
37e4ab3f | 6264 | /* can't change other user's priorities */ |
c69e8d9c | 6265 | if (!check_same_owner(p)) |
37e4ab3f | 6266 | return -EPERM; |
ca94c442 LP |
6267 | |
6268 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6269 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6270 | return -EPERM; | |
37e4ab3f | 6271 | } |
1da177e4 | 6272 | |
725aad24 | 6273 | if (user) { |
b68aa230 | 6274 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6275 | /* |
6276 | * Do not allow realtime tasks into groups that have no runtime | |
6277 | * assigned. | |
6278 | */ | |
9a7e0b18 PZ |
6279 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6280 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6281 | return -EPERM; |
b68aa230 PZ |
6282 | #endif |
6283 | ||
725aad24 JF |
6284 | retval = security_task_setscheduler(p, policy, param); |
6285 | if (retval) | |
6286 | return retval; | |
6287 | } | |
6288 | ||
b29739f9 IM |
6289 | /* |
6290 | * make sure no PI-waiters arrive (or leave) while we are | |
6291 | * changing the priority of the task: | |
6292 | */ | |
6293 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6294 | /* |
6295 | * To be able to change p->policy safely, the apropriate | |
6296 | * runqueue lock must be held. | |
6297 | */ | |
b29739f9 | 6298 | rq = __task_rq_lock(p); |
1da177e4 LT |
6299 | /* recheck policy now with rq lock held */ |
6300 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6301 | policy = oldpolicy = -1; | |
b29739f9 IM |
6302 | __task_rq_unlock(rq); |
6303 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6304 | goto recheck; |
6305 | } | |
2daa3577 | 6306 | update_rq_clock(rq); |
dd41f596 | 6307 | on_rq = p->se.on_rq; |
051a1d1a | 6308 | running = task_current(rq, p); |
0e1f3483 | 6309 | if (on_rq) |
2e1cb74a | 6310 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6311 | if (running) |
6312 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6313 | |
ca94c442 LP |
6314 | p->sched_reset_on_fork = reset_on_fork; |
6315 | ||
1da177e4 | 6316 | oldprio = p->prio; |
dd41f596 | 6317 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6318 | |
0e1f3483 HS |
6319 | if (running) |
6320 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6321 | if (on_rq) { |
6322 | activate_task(rq, p, 0); | |
cb469845 SR |
6323 | |
6324 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6325 | } |
b29739f9 IM |
6326 | __task_rq_unlock(rq); |
6327 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6328 | ||
95e02ca9 TG |
6329 | rt_mutex_adjust_pi(p); |
6330 | ||
1da177e4 LT |
6331 | return 0; |
6332 | } | |
961ccddd RR |
6333 | |
6334 | /** | |
6335 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6336 | * @p: the task in question. | |
6337 | * @policy: new policy. | |
6338 | * @param: structure containing the new RT priority. | |
6339 | * | |
6340 | * NOTE that the task may be already dead. | |
6341 | */ | |
6342 | int sched_setscheduler(struct task_struct *p, int policy, | |
6343 | struct sched_param *param) | |
6344 | { | |
6345 | return __sched_setscheduler(p, policy, param, true); | |
6346 | } | |
1da177e4 LT |
6347 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6348 | ||
961ccddd RR |
6349 | /** |
6350 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6351 | * @p: the task in question. | |
6352 | * @policy: new policy. | |
6353 | * @param: structure containing the new RT priority. | |
6354 | * | |
6355 | * Just like sched_setscheduler, only don't bother checking if the | |
6356 | * current context has permission. For example, this is needed in | |
6357 | * stop_machine(): we create temporary high priority worker threads, | |
6358 | * but our caller might not have that capability. | |
6359 | */ | |
6360 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6361 | struct sched_param *param) | |
6362 | { | |
6363 | return __sched_setscheduler(p, policy, param, false); | |
6364 | } | |
6365 | ||
95cdf3b7 IM |
6366 | static int |
6367 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6368 | { |
1da177e4 LT |
6369 | struct sched_param lparam; |
6370 | struct task_struct *p; | |
36c8b586 | 6371 | int retval; |
1da177e4 LT |
6372 | |
6373 | if (!param || pid < 0) | |
6374 | return -EINVAL; | |
6375 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6376 | return -EFAULT; | |
5fe1d75f ON |
6377 | |
6378 | rcu_read_lock(); | |
6379 | retval = -ESRCH; | |
1da177e4 | 6380 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6381 | if (p != NULL) |
6382 | retval = sched_setscheduler(p, policy, &lparam); | |
6383 | rcu_read_unlock(); | |
36c8b586 | 6384 | |
1da177e4 LT |
6385 | return retval; |
6386 | } | |
6387 | ||
6388 | /** | |
6389 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6390 | * @pid: the pid in question. | |
6391 | * @policy: new policy. | |
6392 | * @param: structure containing the new RT priority. | |
6393 | */ | |
5add95d4 HC |
6394 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6395 | struct sched_param __user *, param) | |
1da177e4 | 6396 | { |
c21761f1 JB |
6397 | /* negative values for policy are not valid */ |
6398 | if (policy < 0) | |
6399 | return -EINVAL; | |
6400 | ||
1da177e4 LT |
6401 | return do_sched_setscheduler(pid, policy, param); |
6402 | } | |
6403 | ||
6404 | /** | |
6405 | * sys_sched_setparam - set/change the RT priority of a thread | |
6406 | * @pid: the pid in question. | |
6407 | * @param: structure containing the new RT priority. | |
6408 | */ | |
5add95d4 | 6409 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6410 | { |
6411 | return do_sched_setscheduler(pid, -1, param); | |
6412 | } | |
6413 | ||
6414 | /** | |
6415 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6416 | * @pid: the pid in question. | |
6417 | */ | |
5add95d4 | 6418 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6419 | { |
36c8b586 | 6420 | struct task_struct *p; |
3a5c359a | 6421 | int retval; |
1da177e4 LT |
6422 | |
6423 | if (pid < 0) | |
3a5c359a | 6424 | return -EINVAL; |
1da177e4 LT |
6425 | |
6426 | retval = -ESRCH; | |
6427 | read_lock(&tasklist_lock); | |
6428 | p = find_process_by_pid(pid); | |
6429 | if (p) { | |
6430 | retval = security_task_getscheduler(p); | |
6431 | if (!retval) | |
ca94c442 LP |
6432 | retval = p->policy |
6433 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 LT |
6434 | } |
6435 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6436 | return retval; |
6437 | } | |
6438 | ||
6439 | /** | |
ca94c442 | 6440 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6441 | * @pid: the pid in question. |
6442 | * @param: structure containing the RT priority. | |
6443 | */ | |
5add95d4 | 6444 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6445 | { |
6446 | struct sched_param lp; | |
36c8b586 | 6447 | struct task_struct *p; |
3a5c359a | 6448 | int retval; |
1da177e4 LT |
6449 | |
6450 | if (!param || pid < 0) | |
3a5c359a | 6451 | return -EINVAL; |
1da177e4 LT |
6452 | |
6453 | read_lock(&tasklist_lock); | |
6454 | p = find_process_by_pid(pid); | |
6455 | retval = -ESRCH; | |
6456 | if (!p) | |
6457 | goto out_unlock; | |
6458 | ||
6459 | retval = security_task_getscheduler(p); | |
6460 | if (retval) | |
6461 | goto out_unlock; | |
6462 | ||
6463 | lp.sched_priority = p->rt_priority; | |
6464 | read_unlock(&tasklist_lock); | |
6465 | ||
6466 | /* | |
6467 | * This one might sleep, we cannot do it with a spinlock held ... | |
6468 | */ | |
6469 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6470 | ||
1da177e4 LT |
6471 | return retval; |
6472 | ||
6473 | out_unlock: | |
6474 | read_unlock(&tasklist_lock); | |
6475 | return retval; | |
6476 | } | |
6477 | ||
96f874e2 | 6478 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6479 | { |
5a16f3d3 | 6480 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6481 | struct task_struct *p; |
6482 | int retval; | |
1da177e4 | 6483 | |
95402b38 | 6484 | get_online_cpus(); |
1da177e4 LT |
6485 | read_lock(&tasklist_lock); |
6486 | ||
6487 | p = find_process_by_pid(pid); | |
6488 | if (!p) { | |
6489 | read_unlock(&tasklist_lock); | |
95402b38 | 6490 | put_online_cpus(); |
1da177e4 LT |
6491 | return -ESRCH; |
6492 | } | |
6493 | ||
6494 | /* | |
6495 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6496 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6497 | * usage count and then drop tasklist_lock. |
6498 | */ | |
6499 | get_task_struct(p); | |
6500 | read_unlock(&tasklist_lock); | |
6501 | ||
5a16f3d3 RR |
6502 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6503 | retval = -ENOMEM; | |
6504 | goto out_put_task; | |
6505 | } | |
6506 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6507 | retval = -ENOMEM; | |
6508 | goto out_free_cpus_allowed; | |
6509 | } | |
1da177e4 | 6510 | retval = -EPERM; |
c69e8d9c | 6511 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6512 | goto out_unlock; |
6513 | ||
e7834f8f DQ |
6514 | retval = security_task_setscheduler(p, 0, NULL); |
6515 | if (retval) | |
6516 | goto out_unlock; | |
6517 | ||
5a16f3d3 RR |
6518 | cpuset_cpus_allowed(p, cpus_allowed); |
6519 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6520 | again: |
5a16f3d3 | 6521 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6522 | |
8707d8b8 | 6523 | if (!retval) { |
5a16f3d3 RR |
6524 | cpuset_cpus_allowed(p, cpus_allowed); |
6525 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6526 | /* |
6527 | * We must have raced with a concurrent cpuset | |
6528 | * update. Just reset the cpus_allowed to the | |
6529 | * cpuset's cpus_allowed | |
6530 | */ | |
5a16f3d3 | 6531 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6532 | goto again; |
6533 | } | |
6534 | } | |
1da177e4 | 6535 | out_unlock: |
5a16f3d3 RR |
6536 | free_cpumask_var(new_mask); |
6537 | out_free_cpus_allowed: | |
6538 | free_cpumask_var(cpus_allowed); | |
6539 | out_put_task: | |
1da177e4 | 6540 | put_task_struct(p); |
95402b38 | 6541 | put_online_cpus(); |
1da177e4 LT |
6542 | return retval; |
6543 | } | |
6544 | ||
6545 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6546 | struct cpumask *new_mask) |
1da177e4 | 6547 | { |
96f874e2 RR |
6548 | if (len < cpumask_size()) |
6549 | cpumask_clear(new_mask); | |
6550 | else if (len > cpumask_size()) | |
6551 | len = cpumask_size(); | |
6552 | ||
1da177e4 LT |
6553 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6554 | } | |
6555 | ||
6556 | /** | |
6557 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6558 | * @pid: pid of the process | |
6559 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6560 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6561 | */ | |
5add95d4 HC |
6562 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6563 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6564 | { |
5a16f3d3 | 6565 | cpumask_var_t new_mask; |
1da177e4 LT |
6566 | int retval; |
6567 | ||
5a16f3d3 RR |
6568 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6569 | return -ENOMEM; | |
1da177e4 | 6570 | |
5a16f3d3 RR |
6571 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6572 | if (retval == 0) | |
6573 | retval = sched_setaffinity(pid, new_mask); | |
6574 | free_cpumask_var(new_mask); | |
6575 | return retval; | |
1da177e4 LT |
6576 | } |
6577 | ||
96f874e2 | 6578 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6579 | { |
36c8b586 | 6580 | struct task_struct *p; |
1da177e4 | 6581 | int retval; |
1da177e4 | 6582 | |
95402b38 | 6583 | get_online_cpus(); |
1da177e4 LT |
6584 | read_lock(&tasklist_lock); |
6585 | ||
6586 | retval = -ESRCH; | |
6587 | p = find_process_by_pid(pid); | |
6588 | if (!p) | |
6589 | goto out_unlock; | |
6590 | ||
e7834f8f DQ |
6591 | retval = security_task_getscheduler(p); |
6592 | if (retval) | |
6593 | goto out_unlock; | |
6594 | ||
96f874e2 | 6595 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6596 | |
6597 | out_unlock: | |
6598 | read_unlock(&tasklist_lock); | |
95402b38 | 6599 | put_online_cpus(); |
1da177e4 | 6600 | |
9531b62f | 6601 | return retval; |
1da177e4 LT |
6602 | } |
6603 | ||
6604 | /** | |
6605 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6606 | * @pid: pid of the process | |
6607 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6608 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6609 | */ | |
5add95d4 HC |
6610 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6611 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6612 | { |
6613 | int ret; | |
f17c8607 | 6614 | cpumask_var_t mask; |
1da177e4 | 6615 | |
f17c8607 | 6616 | if (len < cpumask_size()) |
1da177e4 LT |
6617 | return -EINVAL; |
6618 | ||
f17c8607 RR |
6619 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6620 | return -ENOMEM; | |
1da177e4 | 6621 | |
f17c8607 RR |
6622 | ret = sched_getaffinity(pid, mask); |
6623 | if (ret == 0) { | |
6624 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6625 | ret = -EFAULT; | |
6626 | else | |
6627 | ret = cpumask_size(); | |
6628 | } | |
6629 | free_cpumask_var(mask); | |
1da177e4 | 6630 | |
f17c8607 | 6631 | return ret; |
1da177e4 LT |
6632 | } |
6633 | ||
6634 | /** | |
6635 | * sys_sched_yield - yield the current processor to other threads. | |
6636 | * | |
dd41f596 IM |
6637 | * This function yields the current CPU to other tasks. If there are no |
6638 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6639 | */ |
5add95d4 | 6640 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6641 | { |
70b97a7f | 6642 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6643 | |
2d72376b | 6644 | schedstat_inc(rq, yld_count); |
4530d7ab | 6645 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6646 | |
6647 | /* | |
6648 | * Since we are going to call schedule() anyway, there's | |
6649 | * no need to preempt or enable interrupts: | |
6650 | */ | |
6651 | __release(rq->lock); | |
8a25d5de | 6652 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6653 | _raw_spin_unlock(&rq->lock); |
6654 | preempt_enable_no_resched(); | |
6655 | ||
6656 | schedule(); | |
6657 | ||
6658 | return 0; | |
6659 | } | |
6660 | ||
d86ee480 PZ |
6661 | static inline int should_resched(void) |
6662 | { | |
6663 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6664 | } | |
6665 | ||
e7b38404 | 6666 | static void __cond_resched(void) |
1da177e4 | 6667 | { |
e7aaaa69 FW |
6668 | add_preempt_count(PREEMPT_ACTIVE); |
6669 | schedule(); | |
6670 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6671 | } |
6672 | ||
02b67cc3 | 6673 | int __sched _cond_resched(void) |
1da177e4 | 6674 | { |
d86ee480 | 6675 | if (should_resched()) { |
1da177e4 LT |
6676 | __cond_resched(); |
6677 | return 1; | |
6678 | } | |
6679 | return 0; | |
6680 | } | |
02b67cc3 | 6681 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6682 | |
6683 | /* | |
613afbf8 | 6684 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6685 | * call schedule, and on return reacquire the lock. |
6686 | * | |
41a2d6cf | 6687 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6688 | * operations here to prevent schedule() from being called twice (once via |
6689 | * spin_unlock(), once by hand). | |
6690 | */ | |
613afbf8 | 6691 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6692 | { |
d86ee480 | 6693 | int resched = should_resched(); |
6df3cecb JK |
6694 | int ret = 0; |
6695 | ||
95c354fe | 6696 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6697 | spin_unlock(lock); |
d86ee480 | 6698 | if (resched) |
95c354fe NP |
6699 | __cond_resched(); |
6700 | else | |
6701 | cpu_relax(); | |
6df3cecb | 6702 | ret = 1; |
1da177e4 | 6703 | spin_lock(lock); |
1da177e4 | 6704 | } |
6df3cecb | 6705 | return ret; |
1da177e4 | 6706 | } |
613afbf8 | 6707 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6708 | |
613afbf8 | 6709 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6710 | { |
6711 | BUG_ON(!in_softirq()); | |
6712 | ||
d86ee480 | 6713 | if (should_resched()) { |
98d82567 | 6714 | local_bh_enable(); |
1da177e4 LT |
6715 | __cond_resched(); |
6716 | local_bh_disable(); | |
6717 | return 1; | |
6718 | } | |
6719 | return 0; | |
6720 | } | |
613afbf8 | 6721 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6722 | |
1da177e4 LT |
6723 | /** |
6724 | * yield - yield the current processor to other threads. | |
6725 | * | |
72fd4a35 | 6726 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6727 | * thread runnable and calls sys_sched_yield(). |
6728 | */ | |
6729 | void __sched yield(void) | |
6730 | { | |
6731 | set_current_state(TASK_RUNNING); | |
6732 | sys_sched_yield(); | |
6733 | } | |
1da177e4 LT |
6734 | EXPORT_SYMBOL(yield); |
6735 | ||
6736 | /* | |
41a2d6cf | 6737 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6738 | * that process accounting knows that this is a task in IO wait state. |
6739 | * | |
6740 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6741 | * has set its backing_dev_info: the queue against which it should throttle) | |
6742 | */ | |
6743 | void __sched io_schedule(void) | |
6744 | { | |
54d35f29 | 6745 | struct rq *rq = raw_rq(); |
1da177e4 | 6746 | |
0ff92245 | 6747 | delayacct_blkio_start(); |
1da177e4 LT |
6748 | atomic_inc(&rq->nr_iowait); |
6749 | schedule(); | |
6750 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6751 | delayacct_blkio_end(); |
1da177e4 | 6752 | } |
1da177e4 LT |
6753 | EXPORT_SYMBOL(io_schedule); |
6754 | ||
6755 | long __sched io_schedule_timeout(long timeout) | |
6756 | { | |
54d35f29 | 6757 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6758 | long ret; |
6759 | ||
0ff92245 | 6760 | delayacct_blkio_start(); |
1da177e4 LT |
6761 | atomic_inc(&rq->nr_iowait); |
6762 | ret = schedule_timeout(timeout); | |
6763 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6764 | delayacct_blkio_end(); |
1da177e4 LT |
6765 | return ret; |
6766 | } | |
6767 | ||
6768 | /** | |
6769 | * sys_sched_get_priority_max - return maximum RT priority. | |
6770 | * @policy: scheduling class. | |
6771 | * | |
6772 | * this syscall returns the maximum rt_priority that can be used | |
6773 | * by a given scheduling class. | |
6774 | */ | |
5add95d4 | 6775 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6776 | { |
6777 | int ret = -EINVAL; | |
6778 | ||
6779 | switch (policy) { | |
6780 | case SCHED_FIFO: | |
6781 | case SCHED_RR: | |
6782 | ret = MAX_USER_RT_PRIO-1; | |
6783 | break; | |
6784 | case SCHED_NORMAL: | |
b0a9499c | 6785 | case SCHED_BATCH: |
dd41f596 | 6786 | case SCHED_IDLE: |
1da177e4 LT |
6787 | ret = 0; |
6788 | break; | |
6789 | } | |
6790 | return ret; | |
6791 | } | |
6792 | ||
6793 | /** | |
6794 | * sys_sched_get_priority_min - return minimum RT priority. | |
6795 | * @policy: scheduling class. | |
6796 | * | |
6797 | * this syscall returns the minimum rt_priority that can be used | |
6798 | * by a given scheduling class. | |
6799 | */ | |
5add95d4 | 6800 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6801 | { |
6802 | int ret = -EINVAL; | |
6803 | ||
6804 | switch (policy) { | |
6805 | case SCHED_FIFO: | |
6806 | case SCHED_RR: | |
6807 | ret = 1; | |
6808 | break; | |
6809 | case SCHED_NORMAL: | |
b0a9499c | 6810 | case SCHED_BATCH: |
dd41f596 | 6811 | case SCHED_IDLE: |
1da177e4 LT |
6812 | ret = 0; |
6813 | } | |
6814 | return ret; | |
6815 | } | |
6816 | ||
6817 | /** | |
6818 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6819 | * @pid: pid of the process. | |
6820 | * @interval: userspace pointer to the timeslice value. | |
6821 | * | |
6822 | * this syscall writes the default timeslice value of a given process | |
6823 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6824 | */ | |
17da2bd9 | 6825 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6826 | struct timespec __user *, interval) |
1da177e4 | 6827 | { |
36c8b586 | 6828 | struct task_struct *p; |
a4ec24b4 | 6829 | unsigned int time_slice; |
3a5c359a | 6830 | int retval; |
1da177e4 | 6831 | struct timespec t; |
1da177e4 LT |
6832 | |
6833 | if (pid < 0) | |
3a5c359a | 6834 | return -EINVAL; |
1da177e4 LT |
6835 | |
6836 | retval = -ESRCH; | |
6837 | read_lock(&tasklist_lock); | |
6838 | p = find_process_by_pid(pid); | |
6839 | if (!p) | |
6840 | goto out_unlock; | |
6841 | ||
6842 | retval = security_task_getscheduler(p); | |
6843 | if (retval) | |
6844 | goto out_unlock; | |
6845 | ||
77034937 IM |
6846 | /* |
6847 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6848 | * tasks that are on an otherwise idle runqueue: | |
6849 | */ | |
6850 | time_slice = 0; | |
6851 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6852 | time_slice = DEF_TIMESLICE; |
1868f958 | 6853 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6854 | struct sched_entity *se = &p->se; |
6855 | unsigned long flags; | |
6856 | struct rq *rq; | |
6857 | ||
6858 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6859 | if (rq->cfs.load.weight) |
6860 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6861 | task_rq_unlock(rq, &flags); |
6862 | } | |
1da177e4 | 6863 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6864 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6865 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6866 | return retval; |
3a5c359a | 6867 | |
1da177e4 LT |
6868 | out_unlock: |
6869 | read_unlock(&tasklist_lock); | |
6870 | return retval; | |
6871 | } | |
6872 | ||
7c731e0a | 6873 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6874 | |
82a1fcb9 | 6875 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6876 | { |
1da177e4 | 6877 | unsigned long free = 0; |
36c8b586 | 6878 | unsigned state; |
1da177e4 | 6879 | |
1da177e4 | 6880 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6881 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6882 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6883 | #if BITS_PER_LONG == 32 |
1da177e4 | 6884 | if (state == TASK_RUNNING) |
cc4ea795 | 6885 | printk(KERN_CONT " running "); |
1da177e4 | 6886 | else |
cc4ea795 | 6887 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6888 | #else |
6889 | if (state == TASK_RUNNING) | |
cc4ea795 | 6890 | printk(KERN_CONT " running task "); |
1da177e4 | 6891 | else |
cc4ea795 | 6892 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6893 | #endif |
6894 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6895 | free = stack_not_used(p); |
1da177e4 | 6896 | #endif |
aa47b7e0 DR |
6897 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6898 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6899 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6900 | |
5fb5e6de | 6901 | show_stack(p, NULL); |
1da177e4 LT |
6902 | } |
6903 | ||
e59e2ae2 | 6904 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6905 | { |
36c8b586 | 6906 | struct task_struct *g, *p; |
1da177e4 | 6907 | |
4bd77321 IM |
6908 | #if BITS_PER_LONG == 32 |
6909 | printk(KERN_INFO | |
6910 | " task PC stack pid father\n"); | |
1da177e4 | 6911 | #else |
4bd77321 IM |
6912 | printk(KERN_INFO |
6913 | " task PC stack pid father\n"); | |
1da177e4 LT |
6914 | #endif |
6915 | read_lock(&tasklist_lock); | |
6916 | do_each_thread(g, p) { | |
6917 | /* | |
6918 | * reset the NMI-timeout, listing all files on a slow | |
6919 | * console might take alot of time: | |
6920 | */ | |
6921 | touch_nmi_watchdog(); | |
39bc89fd | 6922 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6923 | sched_show_task(p); |
1da177e4 LT |
6924 | } while_each_thread(g, p); |
6925 | ||
04c9167f JF |
6926 | touch_all_softlockup_watchdogs(); |
6927 | ||
dd41f596 IM |
6928 | #ifdef CONFIG_SCHED_DEBUG |
6929 | sysrq_sched_debug_show(); | |
6930 | #endif | |
1da177e4 | 6931 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6932 | /* |
6933 | * Only show locks if all tasks are dumped: | |
6934 | */ | |
6935 | if (state_filter == -1) | |
6936 | debug_show_all_locks(); | |
1da177e4 LT |
6937 | } |
6938 | ||
1df21055 IM |
6939 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6940 | { | |
dd41f596 | 6941 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6942 | } |
6943 | ||
f340c0d1 IM |
6944 | /** |
6945 | * init_idle - set up an idle thread for a given CPU | |
6946 | * @idle: task in question | |
6947 | * @cpu: cpu the idle task belongs to | |
6948 | * | |
6949 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6950 | * flag, to make booting more robust. | |
6951 | */ | |
5c1e1767 | 6952 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6953 | { |
70b97a7f | 6954 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6955 | unsigned long flags; |
6956 | ||
5cbd54ef IM |
6957 | spin_lock_irqsave(&rq->lock, flags); |
6958 | ||
dd41f596 IM |
6959 | __sched_fork(idle); |
6960 | idle->se.exec_start = sched_clock(); | |
6961 | ||
b29739f9 | 6962 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6963 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6964 | __set_task_cpu(idle, cpu); |
1da177e4 | 6965 | |
1da177e4 | 6966 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6967 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6968 | idle->oncpu = 1; | |
6969 | #endif | |
1da177e4 LT |
6970 | spin_unlock_irqrestore(&rq->lock, flags); |
6971 | ||
6972 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6973 | #if defined(CONFIG_PREEMPT) |
6974 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6975 | #else | |
a1261f54 | 6976 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6977 | #endif |
dd41f596 IM |
6978 | /* |
6979 | * The idle tasks have their own, simple scheduling class: | |
6980 | */ | |
6981 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6982 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6983 | } |
6984 | ||
6985 | /* | |
6986 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6987 | * indicates which cpus entered this state. This is used | |
6988 | * in the rcu update to wait only for active cpus. For system | |
6989 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6990 | * always be CPU_BITS_NONE. |
1da177e4 | 6991 | */ |
6a7b3dc3 | 6992 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6993 | |
19978ca6 IM |
6994 | /* |
6995 | * Increase the granularity value when there are more CPUs, | |
6996 | * because with more CPUs the 'effective latency' as visible | |
6997 | * to users decreases. But the relationship is not linear, | |
6998 | * so pick a second-best guess by going with the log2 of the | |
6999 | * number of CPUs. | |
7000 | * | |
7001 | * This idea comes from the SD scheduler of Con Kolivas: | |
7002 | */ | |
7003 | static inline void sched_init_granularity(void) | |
7004 | { | |
7005 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
7006 | const unsigned long limit = 200000000; | |
7007 | ||
7008 | sysctl_sched_min_granularity *= factor; | |
7009 | if (sysctl_sched_min_granularity > limit) | |
7010 | sysctl_sched_min_granularity = limit; | |
7011 | ||
7012 | sysctl_sched_latency *= factor; | |
7013 | if (sysctl_sched_latency > limit) | |
7014 | sysctl_sched_latency = limit; | |
7015 | ||
7016 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
7017 | |
7018 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
7019 | } |
7020 | ||
1da177e4 LT |
7021 | #ifdef CONFIG_SMP |
7022 | /* | |
7023 | * This is how migration works: | |
7024 | * | |
70b97a7f | 7025 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
7026 | * runqueue and wake up that CPU's migration thread. |
7027 | * 2) we down() the locked semaphore => thread blocks. | |
7028 | * 3) migration thread wakes up (implicitly it forces the migrated | |
7029 | * thread off the CPU) | |
7030 | * 4) it gets the migration request and checks whether the migrated | |
7031 | * task is still in the wrong runqueue. | |
7032 | * 5) if it's in the wrong runqueue then the migration thread removes | |
7033 | * it and puts it into the right queue. | |
7034 | * 6) migration thread up()s the semaphore. | |
7035 | * 7) we wake up and the migration is done. | |
7036 | */ | |
7037 | ||
7038 | /* | |
7039 | * Change a given task's CPU affinity. Migrate the thread to a | |
7040 | * proper CPU and schedule it away if the CPU it's executing on | |
7041 | * is removed from the allowed bitmask. | |
7042 | * | |
7043 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 7044 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
7045 | * call is not atomic; no spinlocks may be held. |
7046 | */ | |
96f874e2 | 7047 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 7048 | { |
70b97a7f | 7049 | struct migration_req req; |
1da177e4 | 7050 | unsigned long flags; |
70b97a7f | 7051 | struct rq *rq; |
48f24c4d | 7052 | int ret = 0; |
1da177e4 LT |
7053 | |
7054 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 7055 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
7056 | ret = -EINVAL; |
7057 | goto out; | |
7058 | } | |
7059 | ||
9985b0ba | 7060 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7061 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7062 | ret = -EINVAL; |
7063 | goto out; | |
7064 | } | |
7065 | ||
73fe6aae | 7066 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7067 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7068 | else { |
96f874e2 RR |
7069 | cpumask_copy(&p->cpus_allowed, new_mask); |
7070 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7071 | } |
7072 | ||
1da177e4 | 7073 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7074 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7075 | goto out; |
7076 | ||
1e5ce4f4 | 7077 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 | 7078 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
7079 | struct task_struct *mt = rq->migration_thread; |
7080 | ||
7081 | get_task_struct(mt); | |
1da177e4 LT |
7082 | task_rq_unlock(rq, &flags); |
7083 | wake_up_process(rq->migration_thread); | |
693525e3 | 7084 | put_task_struct(mt); |
1da177e4 LT |
7085 | wait_for_completion(&req.done); |
7086 | tlb_migrate_finish(p->mm); | |
7087 | return 0; | |
7088 | } | |
7089 | out: | |
7090 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7091 | |
1da177e4 LT |
7092 | return ret; |
7093 | } | |
cd8ba7cd | 7094 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7095 | |
7096 | /* | |
41a2d6cf | 7097 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7098 | * this because either it can't run here any more (set_cpus_allowed() |
7099 | * away from this CPU, or CPU going down), or because we're | |
7100 | * attempting to rebalance this task on exec (sched_exec). | |
7101 | * | |
7102 | * So we race with normal scheduler movements, but that's OK, as long | |
7103 | * as the task is no longer on this CPU. | |
efc30814 KK |
7104 | * |
7105 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7106 | */ |
efc30814 | 7107 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7108 | { |
70b97a7f | 7109 | struct rq *rq_dest, *rq_src; |
dd41f596 | 7110 | int ret = 0, on_rq; |
1da177e4 | 7111 | |
e761b772 | 7112 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7113 | return ret; |
1da177e4 LT |
7114 | |
7115 | rq_src = cpu_rq(src_cpu); | |
7116 | rq_dest = cpu_rq(dest_cpu); | |
7117 | ||
7118 | double_rq_lock(rq_src, rq_dest); | |
7119 | /* Already moved. */ | |
7120 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7121 | goto done; |
1da177e4 | 7122 | /* Affinity changed (again). */ |
96f874e2 | 7123 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7124 | goto fail; |
1da177e4 | 7125 | |
dd41f596 | 7126 | on_rq = p->se.on_rq; |
6e82a3be | 7127 | if (on_rq) |
2e1cb74a | 7128 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7129 | |
1da177e4 | 7130 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7131 | if (on_rq) { |
7132 | activate_task(rq_dest, p, 0); | |
15afe09b | 7133 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7134 | } |
b1e38734 | 7135 | done: |
efc30814 | 7136 | ret = 1; |
b1e38734 | 7137 | fail: |
1da177e4 | 7138 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7139 | return ret; |
1da177e4 LT |
7140 | } |
7141 | ||
7142 | /* | |
7143 | * migration_thread - this is a highprio system thread that performs | |
7144 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7145 | * another runqueue. | |
7146 | */ | |
95cdf3b7 | 7147 | static int migration_thread(void *data) |
1da177e4 | 7148 | { |
1da177e4 | 7149 | int cpu = (long)data; |
70b97a7f | 7150 | struct rq *rq; |
1da177e4 LT |
7151 | |
7152 | rq = cpu_rq(cpu); | |
7153 | BUG_ON(rq->migration_thread != current); | |
7154 | ||
7155 | set_current_state(TASK_INTERRUPTIBLE); | |
7156 | while (!kthread_should_stop()) { | |
70b97a7f | 7157 | struct migration_req *req; |
1da177e4 | 7158 | struct list_head *head; |
1da177e4 | 7159 | |
1da177e4 LT |
7160 | spin_lock_irq(&rq->lock); |
7161 | ||
7162 | if (cpu_is_offline(cpu)) { | |
7163 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7164 | break; |
1da177e4 LT |
7165 | } |
7166 | ||
7167 | if (rq->active_balance) { | |
7168 | active_load_balance(rq, cpu); | |
7169 | rq->active_balance = 0; | |
7170 | } | |
7171 | ||
7172 | head = &rq->migration_queue; | |
7173 | ||
7174 | if (list_empty(head)) { | |
7175 | spin_unlock_irq(&rq->lock); | |
7176 | schedule(); | |
7177 | set_current_state(TASK_INTERRUPTIBLE); | |
7178 | continue; | |
7179 | } | |
70b97a7f | 7180 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7181 | list_del_init(head->next); |
7182 | ||
674311d5 NP |
7183 | spin_unlock(&rq->lock); |
7184 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7185 | local_irq_enable(); | |
1da177e4 LT |
7186 | |
7187 | complete(&req->done); | |
7188 | } | |
7189 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7190 | |
1da177e4 LT |
7191 | return 0; |
7192 | } | |
7193 | ||
7194 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7195 | |
7196 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7197 | { | |
7198 | int ret; | |
7199 | ||
7200 | local_irq_disable(); | |
7201 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7202 | local_irq_enable(); | |
7203 | return ret; | |
7204 | } | |
7205 | ||
054b9108 | 7206 | /* |
3a4fa0a2 | 7207 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7208 | */ |
48f24c4d | 7209 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7210 | { |
70b97a7f | 7211 | int dest_cpu; |
6ca09dfc | 7212 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7213 | |
7214 | again: | |
7215 | /* Look for allowed, online CPU in same node. */ | |
7216 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7217 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7218 | goto move; | |
7219 | ||
7220 | /* Any allowed, online CPU? */ | |
7221 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7222 | if (dest_cpu < nr_cpu_ids) | |
7223 | goto move; | |
7224 | ||
7225 | /* No more Mr. Nice Guy. */ | |
7226 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7227 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7228 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7229 | |
e76bd8d9 RR |
7230 | /* |
7231 | * Don't tell them about moving exiting tasks or | |
7232 | * kernel threads (both mm NULL), since they never | |
7233 | * leave kernel. | |
7234 | */ | |
7235 | if (p->mm && printk_ratelimit()) { | |
7236 | printk(KERN_INFO "process %d (%s) no " | |
7237 | "longer affine to cpu%d\n", | |
7238 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7239 | } |
e76bd8d9 RR |
7240 | } |
7241 | ||
7242 | move: | |
7243 | /* It can have affinity changed while we were choosing. */ | |
7244 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7245 | goto again; | |
1da177e4 LT |
7246 | } |
7247 | ||
7248 | /* | |
7249 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7250 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7251 | * for performance reasons the counter is not stricly tracking tasks to | |
7252 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7253 | * to keep the global sum constant after CPU-down: | |
7254 | */ | |
70b97a7f | 7255 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7256 | { |
1e5ce4f4 | 7257 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7258 | unsigned long flags; |
7259 | ||
7260 | local_irq_save(flags); | |
7261 | double_rq_lock(rq_src, rq_dest); | |
7262 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7263 | rq_src->nr_uninterruptible = 0; | |
7264 | double_rq_unlock(rq_src, rq_dest); | |
7265 | local_irq_restore(flags); | |
7266 | } | |
7267 | ||
7268 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7269 | static void migrate_live_tasks(int src_cpu) | |
7270 | { | |
48f24c4d | 7271 | struct task_struct *p, *t; |
1da177e4 | 7272 | |
f7b4cddc | 7273 | read_lock(&tasklist_lock); |
1da177e4 | 7274 | |
48f24c4d IM |
7275 | do_each_thread(t, p) { |
7276 | if (p == current) | |
1da177e4 LT |
7277 | continue; |
7278 | ||
48f24c4d IM |
7279 | if (task_cpu(p) == src_cpu) |
7280 | move_task_off_dead_cpu(src_cpu, p); | |
7281 | } while_each_thread(t, p); | |
1da177e4 | 7282 | |
f7b4cddc | 7283 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7284 | } |
7285 | ||
dd41f596 IM |
7286 | /* |
7287 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7288 | * It does so by boosting its priority to highest possible. |
7289 | * Used by CPU offline code. | |
1da177e4 LT |
7290 | */ |
7291 | void sched_idle_next(void) | |
7292 | { | |
48f24c4d | 7293 | int this_cpu = smp_processor_id(); |
70b97a7f | 7294 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7295 | struct task_struct *p = rq->idle; |
7296 | unsigned long flags; | |
7297 | ||
7298 | /* cpu has to be offline */ | |
48f24c4d | 7299 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7300 | |
48f24c4d IM |
7301 | /* |
7302 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7303 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7304 | */ |
7305 | spin_lock_irqsave(&rq->lock, flags); | |
7306 | ||
dd41f596 | 7307 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7308 | |
94bc9a7b DA |
7309 | update_rq_clock(rq); |
7310 | activate_task(rq, p, 0); | |
1da177e4 LT |
7311 | |
7312 | spin_unlock_irqrestore(&rq->lock, flags); | |
7313 | } | |
7314 | ||
48f24c4d IM |
7315 | /* |
7316 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7317 | * offline. |
7318 | */ | |
7319 | void idle_task_exit(void) | |
7320 | { | |
7321 | struct mm_struct *mm = current->active_mm; | |
7322 | ||
7323 | BUG_ON(cpu_online(smp_processor_id())); | |
7324 | ||
7325 | if (mm != &init_mm) | |
7326 | switch_mm(mm, &init_mm, current); | |
7327 | mmdrop(mm); | |
7328 | } | |
7329 | ||
054b9108 | 7330 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7331 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7332 | { |
70b97a7f | 7333 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7334 | |
7335 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7336 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7337 | |
7338 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7339 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7340 | |
48f24c4d | 7341 | get_task_struct(p); |
1da177e4 LT |
7342 | |
7343 | /* | |
7344 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7345 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7346 | * fine. |
7347 | */ | |
f7b4cddc | 7348 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7349 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7350 | spin_lock_irq(&rq->lock); |
1da177e4 | 7351 | |
48f24c4d | 7352 | put_task_struct(p); |
1da177e4 LT |
7353 | } |
7354 | ||
7355 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7356 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7357 | { | |
70b97a7f | 7358 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7359 | struct task_struct *next; |
48f24c4d | 7360 | |
dd41f596 IM |
7361 | for ( ; ; ) { |
7362 | if (!rq->nr_running) | |
7363 | break; | |
a8e504d2 | 7364 | update_rq_clock(rq); |
b67802ea | 7365 | next = pick_next_task(rq); |
dd41f596 IM |
7366 | if (!next) |
7367 | break; | |
79c53799 | 7368 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7369 | migrate_dead(dead_cpu, next); |
e692ab53 | 7370 | |
1da177e4 LT |
7371 | } |
7372 | } | |
dce48a84 TG |
7373 | |
7374 | /* | |
7375 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7376 | */ | |
7377 | static void calc_global_load_remove(struct rq *rq) | |
7378 | { | |
7379 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 7380 | rq->calc_load_active = 0; |
dce48a84 | 7381 | } |
1da177e4 LT |
7382 | #endif /* CONFIG_HOTPLUG_CPU */ |
7383 | ||
e692ab53 NP |
7384 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7385 | ||
7386 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7387 | { |
7388 | .procname = "sched_domain", | |
c57baf1e | 7389 | .mode = 0555, |
e0361851 | 7390 | }, |
38605cae | 7391 | {0, }, |
e692ab53 NP |
7392 | }; |
7393 | ||
7394 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7395 | { |
c57baf1e | 7396 | .ctl_name = CTL_KERN, |
e0361851 | 7397 | .procname = "kernel", |
c57baf1e | 7398 | .mode = 0555, |
e0361851 AD |
7399 | .child = sd_ctl_dir, |
7400 | }, | |
38605cae | 7401 | {0, }, |
e692ab53 NP |
7402 | }; |
7403 | ||
7404 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7405 | { | |
7406 | struct ctl_table *entry = | |
5cf9f062 | 7407 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7408 | |
e692ab53 NP |
7409 | return entry; |
7410 | } | |
7411 | ||
6382bc90 MM |
7412 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7413 | { | |
cd790076 | 7414 | struct ctl_table *entry; |
6382bc90 | 7415 | |
cd790076 MM |
7416 | /* |
7417 | * In the intermediate directories, both the child directory and | |
7418 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7419 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7420 | * static strings and all have proc handlers. |
7421 | */ | |
7422 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7423 | if (entry->child) |
7424 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7425 | if (entry->proc_handler == NULL) |
7426 | kfree(entry->procname); | |
7427 | } | |
6382bc90 MM |
7428 | |
7429 | kfree(*tablep); | |
7430 | *tablep = NULL; | |
7431 | } | |
7432 | ||
e692ab53 | 7433 | static void |
e0361851 | 7434 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7435 | const char *procname, void *data, int maxlen, |
7436 | mode_t mode, proc_handler *proc_handler) | |
7437 | { | |
e692ab53 NP |
7438 | entry->procname = procname; |
7439 | entry->data = data; | |
7440 | entry->maxlen = maxlen; | |
7441 | entry->mode = mode; | |
7442 | entry->proc_handler = proc_handler; | |
7443 | } | |
7444 | ||
7445 | static struct ctl_table * | |
7446 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7447 | { | |
a5d8c348 | 7448 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7449 | |
ad1cdc1d MM |
7450 | if (table == NULL) |
7451 | return NULL; | |
7452 | ||
e0361851 | 7453 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7454 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7455 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7456 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7457 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7458 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7459 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7460 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7461 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7462 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7463 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7464 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7465 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7466 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7467 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7468 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7469 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7470 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7471 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7472 | &sd->cache_nice_tries, |
7473 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7474 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7475 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7476 | set_table_entry(&table[11], "name", sd->name, |
7477 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7478 | /* &table[12] is terminator */ | |
e692ab53 NP |
7479 | |
7480 | return table; | |
7481 | } | |
7482 | ||
9a4e7159 | 7483 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7484 | { |
7485 | struct ctl_table *entry, *table; | |
7486 | struct sched_domain *sd; | |
7487 | int domain_num = 0, i; | |
7488 | char buf[32]; | |
7489 | ||
7490 | for_each_domain(cpu, sd) | |
7491 | domain_num++; | |
7492 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7493 | if (table == NULL) |
7494 | return NULL; | |
e692ab53 NP |
7495 | |
7496 | i = 0; | |
7497 | for_each_domain(cpu, sd) { | |
7498 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7499 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7500 | entry->mode = 0555; |
e692ab53 NP |
7501 | entry->child = sd_alloc_ctl_domain_table(sd); |
7502 | entry++; | |
7503 | i++; | |
7504 | } | |
7505 | return table; | |
7506 | } | |
7507 | ||
7508 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7509 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7510 | { |
7511 | int i, cpu_num = num_online_cpus(); | |
7512 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7513 | char buf[32]; | |
7514 | ||
7378547f MM |
7515 | WARN_ON(sd_ctl_dir[0].child); |
7516 | sd_ctl_dir[0].child = entry; | |
7517 | ||
ad1cdc1d MM |
7518 | if (entry == NULL) |
7519 | return; | |
7520 | ||
97b6ea7b | 7521 | for_each_online_cpu(i) { |
e692ab53 | 7522 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7523 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7524 | entry->mode = 0555; |
e692ab53 | 7525 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7526 | entry++; |
e692ab53 | 7527 | } |
7378547f MM |
7528 | |
7529 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7530 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7531 | } | |
6382bc90 | 7532 | |
7378547f | 7533 | /* may be called multiple times per register */ |
6382bc90 MM |
7534 | static void unregister_sched_domain_sysctl(void) |
7535 | { | |
7378547f MM |
7536 | if (sd_sysctl_header) |
7537 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7538 | sd_sysctl_header = NULL; |
7378547f MM |
7539 | if (sd_ctl_dir[0].child) |
7540 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7541 | } |
e692ab53 | 7542 | #else |
6382bc90 MM |
7543 | static void register_sched_domain_sysctl(void) |
7544 | { | |
7545 | } | |
7546 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7547 | { |
7548 | } | |
7549 | #endif | |
7550 | ||
1f11eb6a GH |
7551 | static void set_rq_online(struct rq *rq) |
7552 | { | |
7553 | if (!rq->online) { | |
7554 | const struct sched_class *class; | |
7555 | ||
c6c4927b | 7556 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7557 | rq->online = 1; |
7558 | ||
7559 | for_each_class(class) { | |
7560 | if (class->rq_online) | |
7561 | class->rq_online(rq); | |
7562 | } | |
7563 | } | |
7564 | } | |
7565 | ||
7566 | static void set_rq_offline(struct rq *rq) | |
7567 | { | |
7568 | if (rq->online) { | |
7569 | const struct sched_class *class; | |
7570 | ||
7571 | for_each_class(class) { | |
7572 | if (class->rq_offline) | |
7573 | class->rq_offline(rq); | |
7574 | } | |
7575 | ||
c6c4927b | 7576 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7577 | rq->online = 0; |
7578 | } | |
7579 | } | |
7580 | ||
1da177e4 LT |
7581 | /* |
7582 | * migration_call - callback that gets triggered when a CPU is added. | |
7583 | * Here we can start up the necessary migration thread for the new CPU. | |
7584 | */ | |
48f24c4d IM |
7585 | static int __cpuinit |
7586 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7587 | { |
1da177e4 | 7588 | struct task_struct *p; |
48f24c4d | 7589 | int cpu = (long)hcpu; |
1da177e4 | 7590 | unsigned long flags; |
70b97a7f | 7591 | struct rq *rq; |
1da177e4 LT |
7592 | |
7593 | switch (action) { | |
5be9361c | 7594 | |
1da177e4 | 7595 | case CPU_UP_PREPARE: |
8bb78442 | 7596 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7597 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7598 | if (IS_ERR(p)) |
7599 | return NOTIFY_BAD; | |
1da177e4 LT |
7600 | kthread_bind(p, cpu); |
7601 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7602 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7603 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7604 | task_rq_unlock(rq, &flags); |
371cbb38 | 7605 | get_task_struct(p); |
1da177e4 | 7606 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 7607 | rq->calc_load_update = calc_load_update; |
1da177e4 | 7608 | break; |
48f24c4d | 7609 | |
1da177e4 | 7610 | case CPU_ONLINE: |
8bb78442 | 7611 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7612 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7613 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7614 | |
7615 | /* Update our root-domain */ | |
7616 | rq = cpu_rq(cpu); | |
7617 | spin_lock_irqsave(&rq->lock, flags); | |
7618 | if (rq->rd) { | |
c6c4927b | 7619 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7620 | |
7621 | set_rq_online(rq); | |
1f94ef59 GH |
7622 | } |
7623 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7624 | break; |
48f24c4d | 7625 | |
1da177e4 LT |
7626 | #ifdef CONFIG_HOTPLUG_CPU |
7627 | case CPU_UP_CANCELED: | |
8bb78442 | 7628 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7629 | if (!cpu_rq(cpu)->migration_thread) |
7630 | break; | |
41a2d6cf | 7631 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7632 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7633 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7634 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7635 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7636 | cpu_rq(cpu)->migration_thread = NULL; |
7637 | break; | |
48f24c4d | 7638 | |
1da177e4 | 7639 | case CPU_DEAD: |
8bb78442 | 7640 | case CPU_DEAD_FROZEN: |
470fd646 | 7641 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7642 | migrate_live_tasks(cpu); |
7643 | rq = cpu_rq(cpu); | |
7644 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7645 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7646 | rq->migration_thread = NULL; |
7647 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7648 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7649 | update_rq_clock(rq); |
2e1cb74a | 7650 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7651 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7652 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7653 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7654 | migrate_dead_tasks(cpu); |
d2da272a | 7655 | spin_unlock_irq(&rq->lock); |
470fd646 | 7656 | cpuset_unlock(); |
1da177e4 LT |
7657 | migrate_nr_uninterruptible(rq); |
7658 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7659 | calc_global_load_remove(rq); |
41a2d6cf IM |
7660 | /* |
7661 | * No need to migrate the tasks: it was best-effort if | |
7662 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7663 | * the requestors. | |
7664 | */ | |
1da177e4 LT |
7665 | spin_lock_irq(&rq->lock); |
7666 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7667 | struct migration_req *req; |
7668 | ||
1da177e4 | 7669 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7670 | struct migration_req, list); |
1da177e4 | 7671 | list_del_init(&req->list); |
9a2bd244 | 7672 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7673 | complete(&req->done); |
9a2bd244 | 7674 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7675 | } |
7676 | spin_unlock_irq(&rq->lock); | |
7677 | break; | |
57d885fe | 7678 | |
08f503b0 GH |
7679 | case CPU_DYING: |
7680 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7681 | /* Update our root-domain */ |
7682 | rq = cpu_rq(cpu); | |
7683 | spin_lock_irqsave(&rq->lock, flags); | |
7684 | if (rq->rd) { | |
c6c4927b | 7685 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7686 | set_rq_offline(rq); |
57d885fe GH |
7687 | } |
7688 | spin_unlock_irqrestore(&rq->lock, flags); | |
7689 | break; | |
1da177e4 LT |
7690 | #endif |
7691 | } | |
7692 | return NOTIFY_OK; | |
7693 | } | |
7694 | ||
f38b0820 PM |
7695 | /* |
7696 | * Register at high priority so that task migration (migrate_all_tasks) | |
7697 | * happens before everything else. This has to be lower priority than | |
7698 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7699 | */ |
26c2143b | 7700 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7701 | .notifier_call = migration_call, |
7702 | .priority = 10 | |
7703 | }; | |
7704 | ||
7babe8db | 7705 | static int __init migration_init(void) |
1da177e4 LT |
7706 | { |
7707 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7708 | int err; |
48f24c4d IM |
7709 | |
7710 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7711 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7712 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7713 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7714 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 7715 | |
a004cd42 | 7716 | return 0; |
1da177e4 | 7717 | } |
7babe8db | 7718 | early_initcall(migration_init); |
1da177e4 LT |
7719 | #endif |
7720 | ||
7721 | #ifdef CONFIG_SMP | |
476f3534 | 7722 | |
3e9830dc | 7723 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7724 | |
7c16ec58 | 7725 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7726 | struct cpumask *groupmask) |
1da177e4 | 7727 | { |
4dcf6aff | 7728 | struct sched_group *group = sd->groups; |
434d53b0 | 7729 | char str[256]; |
1da177e4 | 7730 | |
968ea6d8 | 7731 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7732 | cpumask_clear(groupmask); |
4dcf6aff IM |
7733 | |
7734 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7735 | ||
7736 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7737 | printk("does not load-balance\n"); | |
7738 | if (sd->parent) | |
7739 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7740 | " has parent"); | |
7741 | return -1; | |
41c7ce9a NP |
7742 | } |
7743 | ||
eefd796a | 7744 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7745 | |
758b2cdc | 7746 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7747 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7748 | "CPU%d\n", cpu); | |
7749 | } | |
758b2cdc | 7750 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7751 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7752 | " CPU%d\n", cpu); | |
7753 | } | |
1da177e4 | 7754 | |
4dcf6aff | 7755 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7756 | do { |
4dcf6aff IM |
7757 | if (!group) { |
7758 | printk("\n"); | |
7759 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7760 | break; |
7761 | } | |
7762 | ||
4dcf6aff IM |
7763 | if (!group->__cpu_power) { |
7764 | printk(KERN_CONT "\n"); | |
7765 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7766 | "set\n"); | |
7767 | break; | |
7768 | } | |
1da177e4 | 7769 | |
758b2cdc | 7770 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7771 | printk(KERN_CONT "\n"); |
7772 | printk(KERN_ERR "ERROR: empty group\n"); | |
7773 | break; | |
7774 | } | |
1da177e4 | 7775 | |
758b2cdc | 7776 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7777 | printk(KERN_CONT "\n"); |
7778 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7779 | break; | |
7780 | } | |
1da177e4 | 7781 | |
758b2cdc | 7782 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7783 | |
968ea6d8 | 7784 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7785 | |
7786 | printk(KERN_CONT " %s", str); | |
7787 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7788 | printk(KERN_CONT " (__cpu_power = %d)", | |
7789 | group->__cpu_power); | |
7790 | } | |
1da177e4 | 7791 | |
4dcf6aff IM |
7792 | group = group->next; |
7793 | } while (group != sd->groups); | |
7794 | printk(KERN_CONT "\n"); | |
1da177e4 | 7795 | |
758b2cdc | 7796 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7797 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7798 | |
758b2cdc RR |
7799 | if (sd->parent && |
7800 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7801 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7802 | "of domain->span\n"); | |
7803 | return 0; | |
7804 | } | |
1da177e4 | 7805 | |
4dcf6aff IM |
7806 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7807 | { | |
d5dd3db1 | 7808 | cpumask_var_t groupmask; |
4dcf6aff | 7809 | int level = 0; |
1da177e4 | 7810 | |
4dcf6aff IM |
7811 | if (!sd) { |
7812 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7813 | return; | |
7814 | } | |
1da177e4 | 7815 | |
4dcf6aff IM |
7816 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7817 | ||
d5dd3db1 | 7818 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7819 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7820 | return; | |
7821 | } | |
7822 | ||
4dcf6aff | 7823 | for (;;) { |
7c16ec58 | 7824 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7825 | break; |
1da177e4 LT |
7826 | level++; |
7827 | sd = sd->parent; | |
33859f7f | 7828 | if (!sd) |
4dcf6aff IM |
7829 | break; |
7830 | } | |
d5dd3db1 | 7831 | free_cpumask_var(groupmask); |
1da177e4 | 7832 | } |
6d6bc0ad | 7833 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7834 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7835 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7836 | |
1a20ff27 | 7837 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7838 | { |
758b2cdc | 7839 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7840 | return 1; |
7841 | ||
7842 | /* Following flags need at least 2 groups */ | |
7843 | if (sd->flags & (SD_LOAD_BALANCE | | |
7844 | SD_BALANCE_NEWIDLE | | |
7845 | SD_BALANCE_FORK | | |
89c4710e SS |
7846 | SD_BALANCE_EXEC | |
7847 | SD_SHARE_CPUPOWER | | |
7848 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7849 | if (sd->groups != sd->groups->next) |
7850 | return 0; | |
7851 | } | |
7852 | ||
7853 | /* Following flags don't use groups */ | |
7854 | if (sd->flags & (SD_WAKE_IDLE | | |
7855 | SD_WAKE_AFFINE | | |
7856 | SD_WAKE_BALANCE)) | |
7857 | return 0; | |
7858 | ||
7859 | return 1; | |
7860 | } | |
7861 | ||
48f24c4d IM |
7862 | static int |
7863 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7864 | { |
7865 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7866 | ||
7867 | if (sd_degenerate(parent)) | |
7868 | return 1; | |
7869 | ||
758b2cdc | 7870 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7871 | return 0; |
7872 | ||
7873 | /* Does parent contain flags not in child? */ | |
7874 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7875 | if (cflags & SD_WAKE_AFFINE) | |
7876 | pflags &= ~SD_WAKE_BALANCE; | |
7877 | /* Flags needing groups don't count if only 1 group in parent */ | |
7878 | if (parent->groups == parent->groups->next) { | |
7879 | pflags &= ~(SD_LOAD_BALANCE | | |
7880 | SD_BALANCE_NEWIDLE | | |
7881 | SD_BALANCE_FORK | | |
89c4710e SS |
7882 | SD_BALANCE_EXEC | |
7883 | SD_SHARE_CPUPOWER | | |
7884 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7885 | if (nr_node_ids == 1) |
7886 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7887 | } |
7888 | if (~cflags & pflags) | |
7889 | return 0; | |
7890 | ||
7891 | return 1; | |
7892 | } | |
7893 | ||
c6c4927b RR |
7894 | static void free_rootdomain(struct root_domain *rd) |
7895 | { | |
68e74568 RR |
7896 | cpupri_cleanup(&rd->cpupri); |
7897 | ||
c6c4927b RR |
7898 | free_cpumask_var(rd->rto_mask); |
7899 | free_cpumask_var(rd->online); | |
7900 | free_cpumask_var(rd->span); | |
7901 | kfree(rd); | |
7902 | } | |
7903 | ||
57d885fe GH |
7904 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7905 | { | |
a0490fa3 | 7906 | struct root_domain *old_rd = NULL; |
57d885fe | 7907 | unsigned long flags; |
57d885fe GH |
7908 | |
7909 | spin_lock_irqsave(&rq->lock, flags); | |
7910 | ||
7911 | if (rq->rd) { | |
a0490fa3 | 7912 | old_rd = rq->rd; |
57d885fe | 7913 | |
c6c4927b | 7914 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7915 | set_rq_offline(rq); |
57d885fe | 7916 | |
c6c4927b | 7917 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7918 | |
a0490fa3 IM |
7919 | /* |
7920 | * If we dont want to free the old_rt yet then | |
7921 | * set old_rd to NULL to skip the freeing later | |
7922 | * in this function: | |
7923 | */ | |
7924 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7925 | old_rd = NULL; | |
57d885fe GH |
7926 | } |
7927 | ||
7928 | atomic_inc(&rd->refcount); | |
7929 | rq->rd = rd; | |
7930 | ||
c6c4927b | 7931 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 7932 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 7933 | set_rq_online(rq); |
57d885fe GH |
7934 | |
7935 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7936 | |
7937 | if (old_rd) | |
7938 | free_rootdomain(old_rd); | |
57d885fe GH |
7939 | } |
7940 | ||
fd5e1b5d | 7941 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 7942 | { |
36b7b6d4 PE |
7943 | gfp_t gfp = GFP_KERNEL; |
7944 | ||
57d885fe GH |
7945 | memset(rd, 0, sizeof(*rd)); |
7946 | ||
36b7b6d4 PE |
7947 | if (bootmem) |
7948 | gfp = GFP_NOWAIT; | |
c6c4927b | 7949 | |
36b7b6d4 | 7950 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 7951 | goto out; |
36b7b6d4 | 7952 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 7953 | goto free_span; |
36b7b6d4 | 7954 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 7955 | goto free_online; |
6e0534f2 | 7956 | |
0fb53029 | 7957 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 7958 | goto free_rto_mask; |
c6c4927b | 7959 | return 0; |
6e0534f2 | 7960 | |
68e74568 RR |
7961 | free_rto_mask: |
7962 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7963 | free_online: |
7964 | free_cpumask_var(rd->online); | |
7965 | free_span: | |
7966 | free_cpumask_var(rd->span); | |
0c910d28 | 7967 | out: |
c6c4927b | 7968 | return -ENOMEM; |
57d885fe GH |
7969 | } |
7970 | ||
7971 | static void init_defrootdomain(void) | |
7972 | { | |
c6c4927b RR |
7973 | init_rootdomain(&def_root_domain, true); |
7974 | ||
57d885fe GH |
7975 | atomic_set(&def_root_domain.refcount, 1); |
7976 | } | |
7977 | ||
dc938520 | 7978 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7979 | { |
7980 | struct root_domain *rd; | |
7981 | ||
7982 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7983 | if (!rd) | |
7984 | return NULL; | |
7985 | ||
c6c4927b RR |
7986 | if (init_rootdomain(rd, false) != 0) { |
7987 | kfree(rd); | |
7988 | return NULL; | |
7989 | } | |
57d885fe GH |
7990 | |
7991 | return rd; | |
7992 | } | |
7993 | ||
1da177e4 | 7994 | /* |
0eab9146 | 7995 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7996 | * hold the hotplug lock. |
7997 | */ | |
0eab9146 IM |
7998 | static void |
7999 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 8000 | { |
70b97a7f | 8001 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
8002 | struct sched_domain *tmp; |
8003 | ||
8004 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 8005 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
8006 | struct sched_domain *parent = tmp->parent; |
8007 | if (!parent) | |
8008 | break; | |
f29c9b1c | 8009 | |
1a848870 | 8010 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 8011 | tmp->parent = parent->parent; |
1a848870 SS |
8012 | if (parent->parent) |
8013 | parent->parent->child = tmp; | |
f29c9b1c LZ |
8014 | } else |
8015 | tmp = tmp->parent; | |
245af2c7 SS |
8016 | } |
8017 | ||
1a848870 | 8018 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 8019 | sd = sd->parent; |
1a848870 SS |
8020 | if (sd) |
8021 | sd->child = NULL; | |
8022 | } | |
1da177e4 LT |
8023 | |
8024 | sched_domain_debug(sd, cpu); | |
8025 | ||
57d885fe | 8026 | rq_attach_root(rq, rd); |
674311d5 | 8027 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
8028 | } |
8029 | ||
8030 | /* cpus with isolated domains */ | |
dcc30a35 | 8031 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
8032 | |
8033 | /* Setup the mask of cpus configured for isolated domains */ | |
8034 | static int __init isolated_cpu_setup(char *str) | |
8035 | { | |
968ea6d8 | 8036 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
8037 | return 1; |
8038 | } | |
8039 | ||
8927f494 | 8040 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
8041 | |
8042 | /* | |
6711cab4 SS |
8043 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
8044 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
8045 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
8046 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
8047 | * |
8048 | * init_sched_build_groups will build a circular linked list of the groups | |
8049 | * covered by the given span, and will set each group's ->cpumask correctly, | |
8050 | * and ->cpu_power to 0. | |
8051 | */ | |
a616058b | 8052 | static void |
96f874e2 RR |
8053 | init_sched_build_groups(const struct cpumask *span, |
8054 | const struct cpumask *cpu_map, | |
8055 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 8056 | struct sched_group **sg, |
96f874e2 RR |
8057 | struct cpumask *tmpmask), |
8058 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8059 | { |
8060 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8061 | int i; |
8062 | ||
96f874e2 | 8063 | cpumask_clear(covered); |
7c16ec58 | 8064 | |
abcd083a | 8065 | for_each_cpu(i, span) { |
6711cab4 | 8066 | struct sched_group *sg; |
7c16ec58 | 8067 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8068 | int j; |
8069 | ||
758b2cdc | 8070 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8071 | continue; |
8072 | ||
758b2cdc | 8073 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 8074 | sg->__cpu_power = 0; |
1da177e4 | 8075 | |
abcd083a | 8076 | for_each_cpu(j, span) { |
7c16ec58 | 8077 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8078 | continue; |
8079 | ||
96f874e2 | 8080 | cpumask_set_cpu(j, covered); |
758b2cdc | 8081 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8082 | } |
8083 | if (!first) | |
8084 | first = sg; | |
8085 | if (last) | |
8086 | last->next = sg; | |
8087 | last = sg; | |
8088 | } | |
8089 | last->next = first; | |
8090 | } | |
8091 | ||
9c1cfda2 | 8092 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8093 | |
9c1cfda2 | 8094 | #ifdef CONFIG_NUMA |
198e2f18 | 8095 | |
9c1cfda2 JH |
8096 | /** |
8097 | * find_next_best_node - find the next node to include in a sched_domain | |
8098 | * @node: node whose sched_domain we're building | |
8099 | * @used_nodes: nodes already in the sched_domain | |
8100 | * | |
41a2d6cf | 8101 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8102 | * finds the closest node not already in the @used_nodes map. |
8103 | * | |
8104 | * Should use nodemask_t. | |
8105 | */ | |
c5f59f08 | 8106 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8107 | { |
8108 | int i, n, val, min_val, best_node = 0; | |
8109 | ||
8110 | min_val = INT_MAX; | |
8111 | ||
076ac2af | 8112 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8113 | /* Start at @node */ |
076ac2af | 8114 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8115 | |
8116 | if (!nr_cpus_node(n)) | |
8117 | continue; | |
8118 | ||
8119 | /* Skip already used nodes */ | |
c5f59f08 | 8120 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8121 | continue; |
8122 | ||
8123 | /* Simple min distance search */ | |
8124 | val = node_distance(node, n); | |
8125 | ||
8126 | if (val < min_val) { | |
8127 | min_val = val; | |
8128 | best_node = n; | |
8129 | } | |
8130 | } | |
8131 | ||
c5f59f08 | 8132 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8133 | return best_node; |
8134 | } | |
8135 | ||
8136 | /** | |
8137 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8138 | * @node: node whose cpumask we're constructing | |
73486722 | 8139 | * @span: resulting cpumask |
9c1cfda2 | 8140 | * |
41a2d6cf | 8141 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8142 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8143 | * out optimally. | |
8144 | */ | |
96f874e2 | 8145 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8146 | { |
c5f59f08 | 8147 | nodemask_t used_nodes; |
48f24c4d | 8148 | int i; |
9c1cfda2 | 8149 | |
6ca09dfc | 8150 | cpumask_clear(span); |
c5f59f08 | 8151 | nodes_clear(used_nodes); |
9c1cfda2 | 8152 | |
6ca09dfc | 8153 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8154 | node_set(node, used_nodes); |
9c1cfda2 JH |
8155 | |
8156 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8157 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8158 | |
6ca09dfc | 8159 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8160 | } |
9c1cfda2 | 8161 | } |
6d6bc0ad | 8162 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8163 | |
5c45bf27 | 8164 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8165 | |
6c99e9ad RR |
8166 | /* |
8167 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8168 | * |
8169 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8170 | * and struct sched_domain. ) | |
6c99e9ad RR |
8171 | */ |
8172 | struct static_sched_group { | |
8173 | struct sched_group sg; | |
8174 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8175 | }; | |
8176 | ||
8177 | struct static_sched_domain { | |
8178 | struct sched_domain sd; | |
8179 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8180 | }; | |
8181 | ||
9c1cfda2 | 8182 | /* |
48f24c4d | 8183 | * SMT sched-domains: |
9c1cfda2 | 8184 | */ |
1da177e4 | 8185 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8186 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8187 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8188 | |
41a2d6cf | 8189 | static int |
96f874e2 RR |
8190 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8191 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8192 | { |
6711cab4 | 8193 | if (sg) |
6c99e9ad | 8194 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8195 | return cpu; |
8196 | } | |
6d6bc0ad | 8197 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8198 | |
48f24c4d IM |
8199 | /* |
8200 | * multi-core sched-domains: | |
8201 | */ | |
1e9f28fa | 8202 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8203 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8204 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8205 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8206 | |
8207 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8208 | static int |
96f874e2 RR |
8209 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8210 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8211 | { |
6711cab4 | 8212 | int group; |
7c16ec58 | 8213 | |
c69fc56d | 8214 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8215 | group = cpumask_first(mask); |
6711cab4 | 8216 | if (sg) |
6c99e9ad | 8217 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8218 | return group; |
1e9f28fa SS |
8219 | } |
8220 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8221 | static int |
96f874e2 RR |
8222 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8223 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8224 | { |
6711cab4 | 8225 | if (sg) |
6c99e9ad | 8226 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8227 | return cpu; |
8228 | } | |
8229 | #endif | |
8230 | ||
6c99e9ad RR |
8231 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8232 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8233 | |
41a2d6cf | 8234 | static int |
96f874e2 RR |
8235 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8236 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8237 | { |
6711cab4 | 8238 | int group; |
48f24c4d | 8239 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8240 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8241 | group = cpumask_first(mask); |
1e9f28fa | 8242 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8243 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8244 | group = cpumask_first(mask); |
1da177e4 | 8245 | #else |
6711cab4 | 8246 | group = cpu; |
1da177e4 | 8247 | #endif |
6711cab4 | 8248 | if (sg) |
6c99e9ad | 8249 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8250 | return group; |
1da177e4 LT |
8251 | } |
8252 | ||
8253 | #ifdef CONFIG_NUMA | |
1da177e4 | 8254 | /* |
9c1cfda2 JH |
8255 | * The init_sched_build_groups can't handle what we want to do with node |
8256 | * groups, so roll our own. Now each node has its own list of groups which | |
8257 | * gets dynamically allocated. | |
1da177e4 | 8258 | */ |
62ea9ceb | 8259 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8260 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8261 | |
62ea9ceb | 8262 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8263 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8264 | |
96f874e2 RR |
8265 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8266 | struct sched_group **sg, | |
8267 | struct cpumask *nodemask) | |
9c1cfda2 | 8268 | { |
6711cab4 SS |
8269 | int group; |
8270 | ||
6ca09dfc | 8271 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8272 | group = cpumask_first(nodemask); |
6711cab4 SS |
8273 | |
8274 | if (sg) | |
6c99e9ad | 8275 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8276 | return group; |
1da177e4 | 8277 | } |
6711cab4 | 8278 | |
08069033 SS |
8279 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8280 | { | |
8281 | struct sched_group *sg = group_head; | |
8282 | int j; | |
8283 | ||
8284 | if (!sg) | |
8285 | return; | |
3a5c359a | 8286 | do { |
758b2cdc | 8287 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8288 | struct sched_domain *sd; |
08069033 | 8289 | |
6c99e9ad | 8290 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8291 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8292 | /* |
8293 | * Only add "power" once for each | |
8294 | * physical package. | |
8295 | */ | |
8296 | continue; | |
8297 | } | |
08069033 | 8298 | |
3a5c359a AK |
8299 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8300 | } | |
8301 | sg = sg->next; | |
8302 | } while (sg != group_head); | |
08069033 | 8303 | } |
6d6bc0ad | 8304 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8305 | |
a616058b | 8306 | #ifdef CONFIG_NUMA |
51888ca2 | 8307 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8308 | static void free_sched_groups(const struct cpumask *cpu_map, |
8309 | struct cpumask *nodemask) | |
51888ca2 | 8310 | { |
a616058b | 8311 | int cpu, i; |
51888ca2 | 8312 | |
abcd083a | 8313 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8314 | struct sched_group **sched_group_nodes |
8315 | = sched_group_nodes_bycpu[cpu]; | |
8316 | ||
51888ca2 SV |
8317 | if (!sched_group_nodes) |
8318 | continue; | |
8319 | ||
076ac2af | 8320 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8321 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8322 | ||
6ca09dfc | 8323 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8324 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8325 | continue; |
8326 | ||
8327 | if (sg == NULL) | |
8328 | continue; | |
8329 | sg = sg->next; | |
8330 | next_sg: | |
8331 | oldsg = sg; | |
8332 | sg = sg->next; | |
8333 | kfree(oldsg); | |
8334 | if (oldsg != sched_group_nodes[i]) | |
8335 | goto next_sg; | |
8336 | } | |
8337 | kfree(sched_group_nodes); | |
8338 | sched_group_nodes_bycpu[cpu] = NULL; | |
8339 | } | |
51888ca2 | 8340 | } |
6d6bc0ad | 8341 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8342 | static void free_sched_groups(const struct cpumask *cpu_map, |
8343 | struct cpumask *nodemask) | |
a616058b SS |
8344 | { |
8345 | } | |
6d6bc0ad | 8346 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8347 | |
89c4710e SS |
8348 | /* |
8349 | * Initialize sched groups cpu_power. | |
8350 | * | |
8351 | * cpu_power indicates the capacity of sched group, which is used while | |
8352 | * distributing the load between different sched groups in a sched domain. | |
8353 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8354 | * there are asymmetries in the topology. If there are asymmetries, group | |
8355 | * having more cpu_power will pickup more load compared to the group having | |
8356 | * less cpu_power. | |
8357 | * | |
8358 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8359 | * the maximum number of tasks a group can handle in the presence of other idle | |
8360 | * or lightly loaded groups in the same sched domain. | |
8361 | */ | |
8362 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8363 | { | |
8364 | struct sched_domain *child; | |
8365 | struct sched_group *group; | |
8366 | ||
8367 | WARN_ON(!sd || !sd->groups); | |
8368 | ||
13318a71 | 8369 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8370 | return; |
8371 | ||
8372 | child = sd->child; | |
8373 | ||
5517d86b ED |
8374 | sd->groups->__cpu_power = 0; |
8375 | ||
89c4710e SS |
8376 | /* |
8377 | * For perf policy, if the groups in child domain share resources | |
8378 | * (for example cores sharing some portions of the cache hierarchy | |
8379 | * or SMT), then set this domain groups cpu_power such that each group | |
8380 | * can handle only one task, when there are other idle groups in the | |
8381 | * same sched domain. | |
8382 | */ | |
8383 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8384 | (child->flags & | |
8385 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8386 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8387 | return; |
8388 | } | |
8389 | ||
89c4710e SS |
8390 | /* |
8391 | * add cpu_power of each child group to this groups cpu_power | |
8392 | */ | |
8393 | group = child->groups; | |
8394 | do { | |
5517d86b | 8395 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8396 | group = group->next; |
8397 | } while (group != child->groups); | |
8398 | } | |
8399 | ||
7c16ec58 MT |
8400 | /* |
8401 | * Initializers for schedule domains | |
8402 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8403 | */ | |
8404 | ||
a5d8c348 IM |
8405 | #ifdef CONFIG_SCHED_DEBUG |
8406 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8407 | #else | |
8408 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8409 | #endif | |
8410 | ||
7c16ec58 | 8411 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8412 | |
7c16ec58 MT |
8413 | #define SD_INIT_FUNC(type) \ |
8414 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8415 | { \ | |
8416 | memset(sd, 0, sizeof(*sd)); \ | |
8417 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8418 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8419 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8420 | } |
8421 | ||
8422 | SD_INIT_FUNC(CPU) | |
8423 | #ifdef CONFIG_NUMA | |
8424 | SD_INIT_FUNC(ALLNODES) | |
8425 | SD_INIT_FUNC(NODE) | |
8426 | #endif | |
8427 | #ifdef CONFIG_SCHED_SMT | |
8428 | SD_INIT_FUNC(SIBLING) | |
8429 | #endif | |
8430 | #ifdef CONFIG_SCHED_MC | |
8431 | SD_INIT_FUNC(MC) | |
8432 | #endif | |
8433 | ||
1d3504fc HS |
8434 | static int default_relax_domain_level = -1; |
8435 | ||
8436 | static int __init setup_relax_domain_level(char *str) | |
8437 | { | |
30e0e178 LZ |
8438 | unsigned long val; |
8439 | ||
8440 | val = simple_strtoul(str, NULL, 0); | |
8441 | if (val < SD_LV_MAX) | |
8442 | default_relax_domain_level = val; | |
8443 | ||
1d3504fc HS |
8444 | return 1; |
8445 | } | |
8446 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8447 | ||
8448 | static void set_domain_attribute(struct sched_domain *sd, | |
8449 | struct sched_domain_attr *attr) | |
8450 | { | |
8451 | int request; | |
8452 | ||
8453 | if (!attr || attr->relax_domain_level < 0) { | |
8454 | if (default_relax_domain_level < 0) | |
8455 | return; | |
8456 | else | |
8457 | request = default_relax_domain_level; | |
8458 | } else | |
8459 | request = attr->relax_domain_level; | |
8460 | if (request < sd->level) { | |
8461 | /* turn off idle balance on this domain */ | |
8462 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8463 | } else { | |
8464 | /* turn on idle balance on this domain */ | |
8465 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8466 | } | |
8467 | } | |
8468 | ||
1da177e4 | 8469 | /* |
1a20ff27 DG |
8470 | * Build sched domains for a given set of cpus and attach the sched domains |
8471 | * to the individual cpus | |
1da177e4 | 8472 | */ |
96f874e2 | 8473 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8474 | struct sched_domain_attr *attr) |
1da177e4 | 8475 | { |
3404c8d9 | 8476 | int i, err = -ENOMEM; |
57d885fe | 8477 | struct root_domain *rd; |
3404c8d9 RR |
8478 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8479 | tmpmask; | |
d1b55138 | 8480 | #ifdef CONFIG_NUMA |
3404c8d9 | 8481 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8482 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8483 | int sd_allnodes = 0; |
d1b55138 | 8484 | |
3404c8d9 RR |
8485 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8486 | goto out; | |
8487 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8488 | goto free_domainspan; | |
8489 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8490 | goto free_covered; | |
8491 | #endif | |
8492 | ||
8493 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8494 | goto free_notcovered; | |
8495 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8496 | goto free_nodemask; | |
8497 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8498 | goto free_this_sibling_map; | |
8499 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8500 | goto free_this_core_map; | |
8501 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8502 | goto free_send_covered; | |
8503 | ||
8504 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8505 | /* |
8506 | * Allocate the per-node list of sched groups | |
8507 | */ | |
076ac2af | 8508 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8509 | GFP_KERNEL); |
d1b55138 JH |
8510 | if (!sched_group_nodes) { |
8511 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8512 | goto free_tmpmask; |
d1b55138 | 8513 | } |
d1b55138 | 8514 | #endif |
1da177e4 | 8515 | |
dc938520 | 8516 | rd = alloc_rootdomain(); |
57d885fe GH |
8517 | if (!rd) { |
8518 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8519 | goto free_sched_groups; |
57d885fe GH |
8520 | } |
8521 | ||
7c16ec58 | 8522 | #ifdef CONFIG_NUMA |
96f874e2 | 8523 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8524 | #endif |
8525 | ||
1da177e4 | 8526 | /* |
1a20ff27 | 8527 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8528 | */ |
abcd083a | 8529 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8530 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8531 | |
6ca09dfc | 8532 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8533 | |
8534 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8535 | if (cpumask_weight(cpu_map) > |
8536 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8537 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8538 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8539 | set_domain_attribute(sd, attr); |
758b2cdc | 8540 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8541 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8542 | p = sd; |
6711cab4 | 8543 | sd_allnodes = 1; |
9c1cfda2 JH |
8544 | } else |
8545 | p = NULL; | |
8546 | ||
62ea9ceb | 8547 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8548 | SD_INIT(sd, NODE); |
1d3504fc | 8549 | set_domain_attribute(sd, attr); |
758b2cdc | 8550 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8551 | sd->parent = p; |
1a848870 SS |
8552 | if (p) |
8553 | p->child = sd; | |
758b2cdc RR |
8554 | cpumask_and(sched_domain_span(sd), |
8555 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8556 | #endif |
8557 | ||
8558 | p = sd; | |
6c99e9ad | 8559 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8560 | SD_INIT(sd, CPU); |
1d3504fc | 8561 | set_domain_attribute(sd, attr); |
758b2cdc | 8562 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8563 | sd->parent = p; |
1a848870 SS |
8564 | if (p) |
8565 | p->child = sd; | |
7c16ec58 | 8566 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8567 | |
1e9f28fa SS |
8568 | #ifdef CONFIG_SCHED_MC |
8569 | p = sd; | |
6c99e9ad | 8570 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8571 | SD_INIT(sd, MC); |
1d3504fc | 8572 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8573 | cpumask_and(sched_domain_span(sd), cpu_map, |
8574 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8575 | sd->parent = p; |
1a848870 | 8576 | p->child = sd; |
7c16ec58 | 8577 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8578 | #endif |
8579 | ||
1da177e4 LT |
8580 | #ifdef CONFIG_SCHED_SMT |
8581 | p = sd; | |
6c99e9ad | 8582 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8583 | SD_INIT(sd, SIBLING); |
1d3504fc | 8584 | set_domain_attribute(sd, attr); |
758b2cdc | 8585 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8586 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8587 | sd->parent = p; |
1a848870 | 8588 | p->child = sd; |
7c16ec58 | 8589 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8590 | #endif |
8591 | } | |
8592 | ||
8593 | #ifdef CONFIG_SCHED_SMT | |
8594 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8595 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8596 | cpumask_and(this_sibling_map, |
c69fc56d | 8597 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8598 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8599 | continue; |
8600 | ||
dd41f596 | 8601 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8602 | &cpu_to_cpu_group, |
8603 | send_covered, tmpmask); | |
1da177e4 LT |
8604 | } |
8605 | #endif | |
8606 | ||
1e9f28fa SS |
8607 | #ifdef CONFIG_SCHED_MC |
8608 | /* Set up multi-core groups */ | |
abcd083a | 8609 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8610 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8611 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8612 | continue; |
7c16ec58 | 8613 | |
dd41f596 | 8614 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8615 | &cpu_to_core_group, |
8616 | send_covered, tmpmask); | |
1e9f28fa SS |
8617 | } |
8618 | #endif | |
8619 | ||
1da177e4 | 8620 | /* Set up physical groups */ |
076ac2af | 8621 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8622 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8623 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8624 | continue; |
8625 | ||
7c16ec58 MT |
8626 | init_sched_build_groups(nodemask, cpu_map, |
8627 | &cpu_to_phys_group, | |
8628 | send_covered, tmpmask); | |
1da177e4 LT |
8629 | } |
8630 | ||
8631 | #ifdef CONFIG_NUMA | |
8632 | /* Set up node groups */ | |
7c16ec58 | 8633 | if (sd_allnodes) { |
7c16ec58 MT |
8634 | init_sched_build_groups(cpu_map, cpu_map, |
8635 | &cpu_to_allnodes_group, | |
8636 | send_covered, tmpmask); | |
8637 | } | |
9c1cfda2 | 8638 | |
076ac2af | 8639 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8640 | /* Set up node groups */ |
8641 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8642 | int j; |
8643 | ||
96f874e2 | 8644 | cpumask_clear(covered); |
6ca09dfc | 8645 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8646 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8647 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8648 | continue; |
d1b55138 | 8649 | } |
9c1cfda2 | 8650 | |
4bdbaad3 | 8651 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8652 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8653 | |
6c99e9ad RR |
8654 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8655 | GFP_KERNEL, i); | |
51888ca2 SV |
8656 | if (!sg) { |
8657 | printk(KERN_WARNING "Can not alloc domain group for " | |
8658 | "node %d\n", i); | |
8659 | goto error; | |
8660 | } | |
9c1cfda2 | 8661 | sched_group_nodes[i] = sg; |
abcd083a | 8662 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8663 | struct sched_domain *sd; |
9761eea8 | 8664 | |
62ea9ceb | 8665 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8666 | sd->groups = sg; |
9c1cfda2 | 8667 | } |
5517d86b | 8668 | sg->__cpu_power = 0; |
758b2cdc | 8669 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8670 | sg->next = sg; |
96f874e2 | 8671 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8672 | prev = sg; |
8673 | ||
076ac2af | 8674 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8675 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8676 | |
96f874e2 RR |
8677 | cpumask_complement(notcovered, covered); |
8678 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8679 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8680 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8681 | break; |
8682 | ||
6ca09dfc | 8683 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8684 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8685 | continue; |
8686 | ||
6c99e9ad RR |
8687 | sg = kmalloc_node(sizeof(struct sched_group) + |
8688 | cpumask_size(), | |
15f0b676 | 8689 | GFP_KERNEL, i); |
9c1cfda2 JH |
8690 | if (!sg) { |
8691 | printk(KERN_WARNING | |
8692 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8693 | goto error; |
9c1cfda2 | 8694 | } |
5517d86b | 8695 | sg->__cpu_power = 0; |
758b2cdc | 8696 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8697 | sg->next = prev->next; |
96f874e2 | 8698 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8699 | prev->next = sg; |
8700 | prev = sg; | |
8701 | } | |
9c1cfda2 | 8702 | } |
1da177e4 LT |
8703 | #endif |
8704 | ||
8705 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8706 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8707 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8708 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8709 | |
89c4710e | 8710 | init_sched_groups_power(i, sd); |
5c45bf27 | 8711 | } |
1da177e4 | 8712 | #endif |
1e9f28fa | 8713 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8714 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8715 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8716 | |
89c4710e | 8717 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8718 | } |
8719 | #endif | |
1e9f28fa | 8720 | |
abcd083a | 8721 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8722 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8723 | |
89c4710e | 8724 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8725 | } |
8726 | ||
9c1cfda2 | 8727 | #ifdef CONFIG_NUMA |
076ac2af | 8728 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8729 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8730 | |
6711cab4 SS |
8731 | if (sd_allnodes) { |
8732 | struct sched_group *sg; | |
f712c0c7 | 8733 | |
96f874e2 | 8734 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8735 | tmpmask); |
f712c0c7 SS |
8736 | init_numa_sched_groups_power(sg); |
8737 | } | |
9c1cfda2 JH |
8738 | #endif |
8739 | ||
1da177e4 | 8740 | /* Attach the domains */ |
abcd083a | 8741 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8742 | struct sched_domain *sd; |
8743 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8744 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8745 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8746 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8747 | #else |
6c99e9ad | 8748 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8749 | #endif |
57d885fe | 8750 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8751 | } |
51888ca2 | 8752 | |
3404c8d9 RR |
8753 | err = 0; |
8754 | ||
8755 | free_tmpmask: | |
8756 | free_cpumask_var(tmpmask); | |
8757 | free_send_covered: | |
8758 | free_cpumask_var(send_covered); | |
8759 | free_this_core_map: | |
8760 | free_cpumask_var(this_core_map); | |
8761 | free_this_sibling_map: | |
8762 | free_cpumask_var(this_sibling_map); | |
8763 | free_nodemask: | |
8764 | free_cpumask_var(nodemask); | |
8765 | free_notcovered: | |
8766 | #ifdef CONFIG_NUMA | |
8767 | free_cpumask_var(notcovered); | |
8768 | free_covered: | |
8769 | free_cpumask_var(covered); | |
8770 | free_domainspan: | |
8771 | free_cpumask_var(domainspan); | |
8772 | out: | |
8773 | #endif | |
8774 | return err; | |
8775 | ||
8776 | free_sched_groups: | |
8777 | #ifdef CONFIG_NUMA | |
8778 | kfree(sched_group_nodes); | |
8779 | #endif | |
8780 | goto free_tmpmask; | |
51888ca2 | 8781 | |
a616058b | 8782 | #ifdef CONFIG_NUMA |
51888ca2 | 8783 | error: |
7c16ec58 | 8784 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8785 | free_rootdomain(rd); |
3404c8d9 | 8786 | goto free_tmpmask; |
a616058b | 8787 | #endif |
1da177e4 | 8788 | } |
029190c5 | 8789 | |
96f874e2 | 8790 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8791 | { |
8792 | return __build_sched_domains(cpu_map, NULL); | |
8793 | } | |
8794 | ||
96f874e2 | 8795 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8796 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8797 | static struct sched_domain_attr *dattr_cur; |
8798 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8799 | |
8800 | /* | |
8801 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8802 | * cpumask) fails, then fallback to a single sched domain, |
8803 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8804 | */ |
4212823f | 8805 | static cpumask_var_t fallback_doms; |
029190c5 | 8806 | |
ee79d1bd HC |
8807 | /* |
8808 | * arch_update_cpu_topology lets virtualized architectures update the | |
8809 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8810 | * or 0 if it stayed the same. | |
8811 | */ | |
8812 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8813 | { |
ee79d1bd | 8814 | return 0; |
22e52b07 HC |
8815 | } |
8816 | ||
1a20ff27 | 8817 | /* |
41a2d6cf | 8818 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8819 | * For now this just excludes isolated cpus, but could be used to |
8820 | * exclude other special cases in the future. | |
1a20ff27 | 8821 | */ |
96f874e2 | 8822 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8823 | { |
7378547f MM |
8824 | int err; |
8825 | ||
22e52b07 | 8826 | arch_update_cpu_topology(); |
029190c5 | 8827 | ndoms_cur = 1; |
96f874e2 | 8828 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8829 | if (!doms_cur) |
4212823f | 8830 | doms_cur = fallback_doms; |
dcc30a35 | 8831 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8832 | dattr_cur = NULL; |
7378547f | 8833 | err = build_sched_domains(doms_cur); |
6382bc90 | 8834 | register_sched_domain_sysctl(); |
7378547f MM |
8835 | |
8836 | return err; | |
1a20ff27 DG |
8837 | } |
8838 | ||
96f874e2 RR |
8839 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8840 | struct cpumask *tmpmask) | |
1da177e4 | 8841 | { |
7c16ec58 | 8842 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8843 | } |
1da177e4 | 8844 | |
1a20ff27 DG |
8845 | /* |
8846 | * Detach sched domains from a group of cpus specified in cpu_map | |
8847 | * These cpus will now be attached to the NULL domain | |
8848 | */ | |
96f874e2 | 8849 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8850 | { |
96f874e2 RR |
8851 | /* Save because hotplug lock held. */ |
8852 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8853 | int i; |
8854 | ||
abcd083a | 8855 | for_each_cpu(i, cpu_map) |
57d885fe | 8856 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8857 | synchronize_sched(); |
96f874e2 | 8858 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8859 | } |
8860 | ||
1d3504fc HS |
8861 | /* handle null as "default" */ |
8862 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8863 | struct sched_domain_attr *new, int idx_new) | |
8864 | { | |
8865 | struct sched_domain_attr tmp; | |
8866 | ||
8867 | /* fast path */ | |
8868 | if (!new && !cur) | |
8869 | return 1; | |
8870 | ||
8871 | tmp = SD_ATTR_INIT; | |
8872 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8873 | new ? (new + idx_new) : &tmp, | |
8874 | sizeof(struct sched_domain_attr)); | |
8875 | } | |
8876 | ||
029190c5 PJ |
8877 | /* |
8878 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8879 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8880 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8881 | * It destroys each deleted domain and builds each new domain. | |
8882 | * | |
96f874e2 | 8883 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8884 | * The masks don't intersect (don't overlap.) We should setup one |
8885 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8886 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8887 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8888 | * it as it is. | |
8889 | * | |
41a2d6cf IM |
8890 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8891 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8892 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8893 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8894 | * the single partition 'fallback_doms', it also forces the domains | |
8895 | * to be rebuilt. | |
029190c5 | 8896 | * |
96f874e2 | 8897 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8898 | * ndoms_new == 0 is a special case for destroying existing domains, |
8899 | * and it will not create the default domain. | |
dfb512ec | 8900 | * |
029190c5 PJ |
8901 | * Call with hotplug lock held |
8902 | */ | |
96f874e2 RR |
8903 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8904 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8905 | struct sched_domain_attr *dattr_new) |
029190c5 | 8906 | { |
dfb512ec | 8907 | int i, j, n; |
d65bd5ec | 8908 | int new_topology; |
029190c5 | 8909 | |
712555ee | 8910 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8911 | |
7378547f MM |
8912 | /* always unregister in case we don't destroy any domains */ |
8913 | unregister_sched_domain_sysctl(); | |
8914 | ||
d65bd5ec HC |
8915 | /* Let architecture update cpu core mappings. */ |
8916 | new_topology = arch_update_cpu_topology(); | |
8917 | ||
dfb512ec | 8918 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8919 | |
8920 | /* Destroy deleted domains */ | |
8921 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8922 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8923 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8924 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8925 | goto match1; |
8926 | } | |
8927 | /* no match - a current sched domain not in new doms_new[] */ | |
8928 | detach_destroy_domains(doms_cur + i); | |
8929 | match1: | |
8930 | ; | |
8931 | } | |
8932 | ||
e761b772 MK |
8933 | if (doms_new == NULL) { |
8934 | ndoms_cur = 0; | |
4212823f | 8935 | doms_new = fallback_doms; |
dcc30a35 | 8936 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8937 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8938 | } |
8939 | ||
029190c5 PJ |
8940 | /* Build new domains */ |
8941 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8942 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8943 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8944 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8945 | goto match2; |
8946 | } | |
8947 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8948 | __build_sched_domains(doms_new + i, |
8949 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8950 | match2: |
8951 | ; | |
8952 | } | |
8953 | ||
8954 | /* Remember the new sched domains */ | |
4212823f | 8955 | if (doms_cur != fallback_doms) |
029190c5 | 8956 | kfree(doms_cur); |
1d3504fc | 8957 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8958 | doms_cur = doms_new; |
1d3504fc | 8959 | dattr_cur = dattr_new; |
029190c5 | 8960 | ndoms_cur = ndoms_new; |
7378547f MM |
8961 | |
8962 | register_sched_domain_sysctl(); | |
a1835615 | 8963 | |
712555ee | 8964 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8965 | } |
8966 | ||
5c45bf27 | 8967 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8968 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8969 | { |
95402b38 | 8970 | get_online_cpus(); |
dfb512ec MK |
8971 | |
8972 | /* Destroy domains first to force the rebuild */ | |
8973 | partition_sched_domains(0, NULL, NULL); | |
8974 | ||
e761b772 | 8975 | rebuild_sched_domains(); |
95402b38 | 8976 | put_online_cpus(); |
5c45bf27 SS |
8977 | } |
8978 | ||
8979 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8980 | { | |
afb8a9b7 | 8981 | unsigned int level = 0; |
5c45bf27 | 8982 | |
afb8a9b7 GS |
8983 | if (sscanf(buf, "%u", &level) != 1) |
8984 | return -EINVAL; | |
8985 | ||
8986 | /* | |
8987 | * level is always be positive so don't check for | |
8988 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8989 | * What happens on 0 or 1 byte write, | |
8990 | * need to check for count as well? | |
8991 | */ | |
8992 | ||
8993 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8994 | return -EINVAL; |
8995 | ||
8996 | if (smt) | |
afb8a9b7 | 8997 | sched_smt_power_savings = level; |
5c45bf27 | 8998 | else |
afb8a9b7 | 8999 | sched_mc_power_savings = level; |
5c45bf27 | 9000 | |
c70f22d2 | 9001 | arch_reinit_sched_domains(); |
5c45bf27 | 9002 | |
c70f22d2 | 9003 | return count; |
5c45bf27 SS |
9004 | } |
9005 | ||
5c45bf27 | 9006 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
9007 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
9008 | char *page) | |
5c45bf27 SS |
9009 | { |
9010 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
9011 | } | |
f718cd4a | 9012 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 9013 | const char *buf, size_t count) |
5c45bf27 SS |
9014 | { |
9015 | return sched_power_savings_store(buf, count, 0); | |
9016 | } | |
f718cd4a AK |
9017 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
9018 | sched_mc_power_savings_show, | |
9019 | sched_mc_power_savings_store); | |
5c45bf27 SS |
9020 | #endif |
9021 | ||
9022 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
9023 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
9024 | char *page) | |
5c45bf27 SS |
9025 | { |
9026 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
9027 | } | |
f718cd4a | 9028 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 9029 | const char *buf, size_t count) |
5c45bf27 SS |
9030 | { |
9031 | return sched_power_savings_store(buf, count, 1); | |
9032 | } | |
f718cd4a AK |
9033 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
9034 | sched_smt_power_savings_show, | |
6707de00 AB |
9035 | sched_smt_power_savings_store); |
9036 | #endif | |
9037 | ||
39aac648 | 9038 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
9039 | { |
9040 | int err = 0; | |
9041 | ||
9042 | #ifdef CONFIG_SCHED_SMT | |
9043 | if (smt_capable()) | |
9044 | err = sysfs_create_file(&cls->kset.kobj, | |
9045 | &attr_sched_smt_power_savings.attr); | |
9046 | #endif | |
9047 | #ifdef CONFIG_SCHED_MC | |
9048 | if (!err && mc_capable()) | |
9049 | err = sysfs_create_file(&cls->kset.kobj, | |
9050 | &attr_sched_mc_power_savings.attr); | |
9051 | #endif | |
9052 | return err; | |
9053 | } | |
6d6bc0ad | 9054 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 9055 | |
e761b772 | 9056 | #ifndef CONFIG_CPUSETS |
1da177e4 | 9057 | /* |
e761b772 MK |
9058 | * Add online and remove offline CPUs from the scheduler domains. |
9059 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9060 | */ |
9061 | static int update_sched_domains(struct notifier_block *nfb, | |
9062 | unsigned long action, void *hcpu) | |
e761b772 MK |
9063 | { |
9064 | switch (action) { | |
9065 | case CPU_ONLINE: | |
9066 | case CPU_ONLINE_FROZEN: | |
9067 | case CPU_DEAD: | |
9068 | case CPU_DEAD_FROZEN: | |
dfb512ec | 9069 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9070 | return NOTIFY_OK; |
9071 | ||
9072 | default: | |
9073 | return NOTIFY_DONE; | |
9074 | } | |
9075 | } | |
9076 | #endif | |
9077 | ||
9078 | static int update_runtime(struct notifier_block *nfb, | |
9079 | unsigned long action, void *hcpu) | |
1da177e4 | 9080 | { |
7def2be1 PZ |
9081 | int cpu = (int)(long)hcpu; |
9082 | ||
1da177e4 | 9083 | switch (action) { |
1da177e4 | 9084 | case CPU_DOWN_PREPARE: |
8bb78442 | 9085 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9086 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9087 | return NOTIFY_OK; |
9088 | ||
1da177e4 | 9089 | case CPU_DOWN_FAILED: |
8bb78442 | 9090 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9091 | case CPU_ONLINE: |
8bb78442 | 9092 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9093 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9094 | return NOTIFY_OK; |
9095 | ||
1da177e4 LT |
9096 | default: |
9097 | return NOTIFY_DONE; | |
9098 | } | |
1da177e4 | 9099 | } |
1da177e4 LT |
9100 | |
9101 | void __init sched_init_smp(void) | |
9102 | { | |
dcc30a35 RR |
9103 | cpumask_var_t non_isolated_cpus; |
9104 | ||
9105 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 9106 | |
434d53b0 MT |
9107 | #if defined(CONFIG_NUMA) |
9108 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9109 | GFP_KERNEL); | |
9110 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9111 | #endif | |
95402b38 | 9112 | get_online_cpus(); |
712555ee | 9113 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9114 | arch_init_sched_domains(cpu_online_mask); |
9115 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9116 | if (cpumask_empty(non_isolated_cpus)) | |
9117 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9118 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9119 | put_online_cpus(); |
e761b772 MK |
9120 | |
9121 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9122 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9123 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9124 | #endif |
9125 | ||
9126 | /* RT runtime code needs to handle some hotplug events */ | |
9127 | hotcpu_notifier(update_runtime, 0); | |
9128 | ||
b328ca18 | 9129 | init_hrtick(); |
5c1e1767 NP |
9130 | |
9131 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9132 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9133 | BUG(); |
19978ca6 | 9134 | sched_init_granularity(); |
dcc30a35 | 9135 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9136 | |
9137 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9138 | init_sched_rt_class(); |
1da177e4 LT |
9139 | } |
9140 | #else | |
9141 | void __init sched_init_smp(void) | |
9142 | { | |
19978ca6 | 9143 | sched_init_granularity(); |
1da177e4 LT |
9144 | } |
9145 | #endif /* CONFIG_SMP */ | |
9146 | ||
cd1bb94b AB |
9147 | const_debug unsigned int sysctl_timer_migration = 1; |
9148 | ||
1da177e4 LT |
9149 | int in_sched_functions(unsigned long addr) |
9150 | { | |
1da177e4 LT |
9151 | return in_lock_functions(addr) || |
9152 | (addr >= (unsigned long)__sched_text_start | |
9153 | && addr < (unsigned long)__sched_text_end); | |
9154 | } | |
9155 | ||
a9957449 | 9156 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9157 | { |
9158 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9159 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9160 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9161 | cfs_rq->rq = rq; | |
9162 | #endif | |
67e9fb2a | 9163 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9164 | } |
9165 | ||
fa85ae24 PZ |
9166 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9167 | { | |
9168 | struct rt_prio_array *array; | |
9169 | int i; | |
9170 | ||
9171 | array = &rt_rq->active; | |
9172 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9173 | INIT_LIST_HEAD(array->queue + i); | |
9174 | __clear_bit(i, array->bitmap); | |
9175 | } | |
9176 | /* delimiter for bitsearch: */ | |
9177 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9178 | ||
052f1dc7 | 9179 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9180 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9181 | #ifdef CONFIG_SMP |
e864c499 | 9182 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9183 | #endif |
48d5e258 | 9184 | #endif |
fa85ae24 PZ |
9185 | #ifdef CONFIG_SMP |
9186 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9187 | rt_rq->overloaded = 0; |
c20b08e3 | 9188 | plist_head_init(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9189 | #endif |
9190 | ||
9191 | rt_rq->rt_time = 0; | |
9192 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9193 | rt_rq->rt_runtime = 0; |
9194 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9195 | |
052f1dc7 | 9196 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9197 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9198 | rt_rq->rq = rq; |
9199 | #endif | |
fa85ae24 PZ |
9200 | } |
9201 | ||
6f505b16 | 9202 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9203 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9204 | struct sched_entity *se, int cpu, int add, | |
9205 | struct sched_entity *parent) | |
6f505b16 | 9206 | { |
ec7dc8ac | 9207 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9208 | tg->cfs_rq[cpu] = cfs_rq; |
9209 | init_cfs_rq(cfs_rq, rq); | |
9210 | cfs_rq->tg = tg; | |
9211 | if (add) | |
9212 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9213 | ||
9214 | tg->se[cpu] = se; | |
354d60c2 DG |
9215 | /* se could be NULL for init_task_group */ |
9216 | if (!se) | |
9217 | return; | |
9218 | ||
ec7dc8ac DG |
9219 | if (!parent) |
9220 | se->cfs_rq = &rq->cfs; | |
9221 | else | |
9222 | se->cfs_rq = parent->my_q; | |
9223 | ||
6f505b16 PZ |
9224 | se->my_q = cfs_rq; |
9225 | se->load.weight = tg->shares; | |
e05510d0 | 9226 | se->load.inv_weight = 0; |
ec7dc8ac | 9227 | se->parent = parent; |
6f505b16 | 9228 | } |
052f1dc7 | 9229 | #endif |
6f505b16 | 9230 | |
052f1dc7 | 9231 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9232 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9233 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9234 | struct sched_rt_entity *parent) | |
6f505b16 | 9235 | { |
ec7dc8ac DG |
9236 | struct rq *rq = cpu_rq(cpu); |
9237 | ||
6f505b16 PZ |
9238 | tg->rt_rq[cpu] = rt_rq; |
9239 | init_rt_rq(rt_rq, rq); | |
9240 | rt_rq->tg = tg; | |
9241 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9242 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9243 | if (add) |
9244 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9245 | ||
9246 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9247 | if (!rt_se) |
9248 | return; | |
9249 | ||
ec7dc8ac DG |
9250 | if (!parent) |
9251 | rt_se->rt_rq = &rq->rt; | |
9252 | else | |
9253 | rt_se->rt_rq = parent->my_q; | |
9254 | ||
6f505b16 | 9255 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9256 | rt_se->parent = parent; |
6f505b16 PZ |
9257 | INIT_LIST_HEAD(&rt_se->run_list); |
9258 | } | |
9259 | #endif | |
9260 | ||
1da177e4 LT |
9261 | void __init sched_init(void) |
9262 | { | |
dd41f596 | 9263 | int i, j; |
434d53b0 MT |
9264 | unsigned long alloc_size = 0, ptr; |
9265 | ||
9266 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9267 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9268 | #endif | |
9269 | #ifdef CONFIG_RT_GROUP_SCHED | |
9270 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9271 | #endif |
9272 | #ifdef CONFIG_USER_SCHED | |
9273 | alloc_size *= 2; | |
df7c8e84 RR |
9274 | #endif |
9275 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9276 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9277 | #endif |
9278 | /* | |
9279 | * As sched_init() is called before page_alloc is setup, | |
9280 | * we use alloc_bootmem(). | |
9281 | */ | |
9282 | if (alloc_size) { | |
36b7b6d4 | 9283 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9284 | |
9285 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9286 | init_task_group.se = (struct sched_entity **)ptr; | |
9287 | ptr += nr_cpu_ids * sizeof(void **); | |
9288 | ||
9289 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9290 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9291 | |
9292 | #ifdef CONFIG_USER_SCHED | |
9293 | root_task_group.se = (struct sched_entity **)ptr; | |
9294 | ptr += nr_cpu_ids * sizeof(void **); | |
9295 | ||
9296 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9297 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9298 | #endif /* CONFIG_USER_SCHED */ |
9299 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9300 | #ifdef CONFIG_RT_GROUP_SCHED |
9301 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9302 | ptr += nr_cpu_ids * sizeof(void **); | |
9303 | ||
9304 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9305 | ptr += nr_cpu_ids * sizeof(void **); |
9306 | ||
9307 | #ifdef CONFIG_USER_SCHED | |
9308 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9309 | ptr += nr_cpu_ids * sizeof(void **); | |
9310 | ||
9311 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9312 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9313 | #endif /* CONFIG_USER_SCHED */ |
9314 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9315 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9316 | for_each_possible_cpu(i) { | |
9317 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9318 | ptr += cpumask_size(); | |
9319 | } | |
9320 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9321 | } |
dd41f596 | 9322 | |
57d885fe GH |
9323 | #ifdef CONFIG_SMP |
9324 | init_defrootdomain(); | |
9325 | #endif | |
9326 | ||
d0b27fa7 PZ |
9327 | init_rt_bandwidth(&def_rt_bandwidth, |
9328 | global_rt_period(), global_rt_runtime()); | |
9329 | ||
9330 | #ifdef CONFIG_RT_GROUP_SCHED | |
9331 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9332 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9333 | #ifdef CONFIG_USER_SCHED |
9334 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9335 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9336 | #endif /* CONFIG_USER_SCHED */ |
9337 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9338 | |
052f1dc7 | 9339 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9340 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9341 | INIT_LIST_HEAD(&init_task_group.children); |
9342 | ||
9343 | #ifdef CONFIG_USER_SCHED | |
9344 | INIT_LIST_HEAD(&root_task_group.children); | |
9345 | init_task_group.parent = &root_task_group; | |
9346 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9347 | #endif /* CONFIG_USER_SCHED */ |
9348 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9349 | |
0a945022 | 9350 | for_each_possible_cpu(i) { |
70b97a7f | 9351 | struct rq *rq; |
1da177e4 LT |
9352 | |
9353 | rq = cpu_rq(i); | |
9354 | spin_lock_init(&rq->lock); | |
7897986b | 9355 | rq->nr_running = 0; |
dce48a84 TG |
9356 | rq->calc_load_active = 0; |
9357 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9358 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9359 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9360 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9361 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9362 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9363 | #ifdef CONFIG_CGROUP_SCHED |
9364 | /* | |
9365 | * How much cpu bandwidth does init_task_group get? | |
9366 | * | |
9367 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9368 | * gets 100% of the cpu resources in the system. This overall | |
9369 | * system cpu resource is divided among the tasks of | |
9370 | * init_task_group and its child task-groups in a fair manner, | |
9371 | * based on each entity's (task or task-group's) weight | |
9372 | * (se->load.weight). | |
9373 | * | |
9374 | * In other words, if init_task_group has 10 tasks of weight | |
9375 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9376 | * then A0's share of the cpu resource is: | |
9377 | * | |
0d905bca | 9378 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9379 | * |
9380 | * We achieve this by letting init_task_group's tasks sit | |
9381 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9382 | */ | |
ec7dc8ac | 9383 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9384 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9385 | root_task_group.shares = NICE_0_LOAD; |
9386 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9387 | /* |
9388 | * In case of task-groups formed thr' the user id of tasks, | |
9389 | * init_task_group represents tasks belonging to root user. | |
9390 | * Hence it forms a sibling of all subsequent groups formed. | |
9391 | * In this case, init_task_group gets only a fraction of overall | |
9392 | * system cpu resource, based on the weight assigned to root | |
9393 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9394 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9395 | * (init_cfs_rq) and having one entity represent this group of | |
9396 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9397 | */ | |
ec7dc8ac | 9398 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9399 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9400 | &per_cpu(init_sched_entity, i), i, 1, |
9401 | root_task_group.se[i]); | |
6f505b16 | 9402 | |
052f1dc7 | 9403 | #endif |
354d60c2 DG |
9404 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9405 | ||
9406 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9407 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9408 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9409 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9410 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9411 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9412 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9413 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9414 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9415 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9416 | root_task_group.rt_se[i]); | |
354d60c2 | 9417 | #endif |
dd41f596 | 9418 | #endif |
1da177e4 | 9419 | |
dd41f596 IM |
9420 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9421 | rq->cpu_load[j] = 0; | |
1da177e4 | 9422 | #ifdef CONFIG_SMP |
41c7ce9a | 9423 | rq->sd = NULL; |
57d885fe | 9424 | rq->rd = NULL; |
3f029d3c | 9425 | rq->post_schedule = 0; |
1da177e4 | 9426 | rq->active_balance = 0; |
dd41f596 | 9427 | rq->next_balance = jiffies; |
1da177e4 | 9428 | rq->push_cpu = 0; |
0a2966b4 | 9429 | rq->cpu = i; |
1f11eb6a | 9430 | rq->online = 0; |
1da177e4 LT |
9431 | rq->migration_thread = NULL; |
9432 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9433 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9434 | #endif |
8f4d37ec | 9435 | init_rq_hrtick(rq); |
1da177e4 | 9436 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9437 | } |
9438 | ||
2dd73a4f | 9439 | set_load_weight(&init_task); |
b50f60ce | 9440 | |
e107be36 AK |
9441 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9442 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9443 | #endif | |
9444 | ||
c9819f45 | 9445 | #ifdef CONFIG_SMP |
962cf36c | 9446 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9447 | #endif |
9448 | ||
b50f60ce HC |
9449 | #ifdef CONFIG_RT_MUTEXES |
9450 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9451 | #endif | |
9452 | ||
1da177e4 LT |
9453 | /* |
9454 | * The boot idle thread does lazy MMU switching as well: | |
9455 | */ | |
9456 | atomic_inc(&init_mm.mm_count); | |
9457 | enter_lazy_tlb(&init_mm, current); | |
9458 | ||
9459 | /* | |
9460 | * Make us the idle thread. Technically, schedule() should not be | |
9461 | * called from this thread, however somewhere below it might be, | |
9462 | * but because we are the idle thread, we just pick up running again | |
9463 | * when this runqueue becomes "idle". | |
9464 | */ | |
9465 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9466 | |
9467 | calc_load_update = jiffies + LOAD_FREQ; | |
9468 | ||
dd41f596 IM |
9469 | /* |
9470 | * During early bootup we pretend to be a normal task: | |
9471 | */ | |
9472 | current->sched_class = &fair_sched_class; | |
6892b75e | 9473 | |
6a7b3dc3 | 9474 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9475 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9476 | #ifdef CONFIG_SMP |
7d1e6a9b | 9477 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9478 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9479 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9480 | #endif |
4bdddf8f | 9481 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9482 | #endif /* SMP */ |
6a7b3dc3 | 9483 | |
0d905bca IM |
9484 | perf_counter_init(); |
9485 | ||
6892b75e | 9486 | scheduler_running = 1; |
1da177e4 LT |
9487 | } |
9488 | ||
9489 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9490 | static inline int preempt_count_equals(int preempt_offset) |
9491 | { | |
9492 | int nested = preempt_count() & ~PREEMPT_ACTIVE; | |
9493 | ||
9494 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9495 | } | |
9496 | ||
9497 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9498 | { |
48f24c4d | 9499 | #ifdef in_atomic |
1da177e4 LT |
9500 | static unsigned long prev_jiffy; /* ratelimiting */ |
9501 | ||
e4aafea2 FW |
9502 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9503 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9504 | return; |
9505 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9506 | return; | |
9507 | prev_jiffy = jiffies; | |
9508 | ||
9509 | printk(KERN_ERR | |
9510 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9511 | file, line); | |
9512 | printk(KERN_ERR | |
9513 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9514 | in_atomic(), irqs_disabled(), | |
9515 | current->pid, current->comm); | |
9516 | ||
9517 | debug_show_held_locks(current); | |
9518 | if (irqs_disabled()) | |
9519 | print_irqtrace_events(current); | |
9520 | dump_stack(); | |
1da177e4 LT |
9521 | #endif |
9522 | } | |
9523 | EXPORT_SYMBOL(__might_sleep); | |
9524 | #endif | |
9525 | ||
9526 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9527 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9528 | { | |
9529 | int on_rq; | |
3e51f33f | 9530 | |
3a5e4dc1 AK |
9531 | update_rq_clock(rq); |
9532 | on_rq = p->se.on_rq; | |
9533 | if (on_rq) | |
9534 | deactivate_task(rq, p, 0); | |
9535 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9536 | if (on_rq) { | |
9537 | activate_task(rq, p, 0); | |
9538 | resched_task(rq->curr); | |
9539 | } | |
9540 | } | |
9541 | ||
1da177e4 LT |
9542 | void normalize_rt_tasks(void) |
9543 | { | |
a0f98a1c | 9544 | struct task_struct *g, *p; |
1da177e4 | 9545 | unsigned long flags; |
70b97a7f | 9546 | struct rq *rq; |
1da177e4 | 9547 | |
4cf5d77a | 9548 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9549 | do_each_thread(g, p) { |
178be793 IM |
9550 | /* |
9551 | * Only normalize user tasks: | |
9552 | */ | |
9553 | if (!p->mm) | |
9554 | continue; | |
9555 | ||
6cfb0d5d | 9556 | p->se.exec_start = 0; |
6cfb0d5d | 9557 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9558 | p->se.wait_start = 0; |
dd41f596 | 9559 | p->se.sleep_start = 0; |
dd41f596 | 9560 | p->se.block_start = 0; |
6cfb0d5d | 9561 | #endif |
dd41f596 IM |
9562 | |
9563 | if (!rt_task(p)) { | |
9564 | /* | |
9565 | * Renice negative nice level userspace | |
9566 | * tasks back to 0: | |
9567 | */ | |
9568 | if (TASK_NICE(p) < 0 && p->mm) | |
9569 | set_user_nice(p, 0); | |
1da177e4 | 9570 | continue; |
dd41f596 | 9571 | } |
1da177e4 | 9572 | |
4cf5d77a | 9573 | spin_lock(&p->pi_lock); |
b29739f9 | 9574 | rq = __task_rq_lock(p); |
1da177e4 | 9575 | |
178be793 | 9576 | normalize_task(rq, p); |
3a5e4dc1 | 9577 | |
b29739f9 | 9578 | __task_rq_unlock(rq); |
4cf5d77a | 9579 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9580 | } while_each_thread(g, p); |
9581 | ||
4cf5d77a | 9582 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9583 | } |
9584 | ||
9585 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9586 | |
9587 | #ifdef CONFIG_IA64 | |
9588 | /* | |
9589 | * These functions are only useful for the IA64 MCA handling. | |
9590 | * | |
9591 | * They can only be called when the whole system has been | |
9592 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9593 | * activity can take place. Using them for anything else would | |
9594 | * be a serious bug, and as a result, they aren't even visible | |
9595 | * under any other configuration. | |
9596 | */ | |
9597 | ||
9598 | /** | |
9599 | * curr_task - return the current task for a given cpu. | |
9600 | * @cpu: the processor in question. | |
9601 | * | |
9602 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9603 | */ | |
36c8b586 | 9604 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9605 | { |
9606 | return cpu_curr(cpu); | |
9607 | } | |
9608 | ||
9609 | /** | |
9610 | * set_curr_task - set the current task for a given cpu. | |
9611 | * @cpu: the processor in question. | |
9612 | * @p: the task pointer to set. | |
9613 | * | |
9614 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9615 | * are serviced on a separate stack. It allows the architecture to switch the |
9616 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9617 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9618 | * and caller must save the original value of the current task (see | |
9619 | * curr_task() above) and restore that value before reenabling interrupts and | |
9620 | * re-starting the system. | |
9621 | * | |
9622 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9623 | */ | |
36c8b586 | 9624 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9625 | { |
9626 | cpu_curr(cpu) = p; | |
9627 | } | |
9628 | ||
9629 | #endif | |
29f59db3 | 9630 | |
bccbe08a PZ |
9631 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9632 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9633 | { |
9634 | int i; | |
9635 | ||
9636 | for_each_possible_cpu(i) { | |
9637 | if (tg->cfs_rq) | |
9638 | kfree(tg->cfs_rq[i]); | |
9639 | if (tg->se) | |
9640 | kfree(tg->se[i]); | |
6f505b16 PZ |
9641 | } |
9642 | ||
9643 | kfree(tg->cfs_rq); | |
9644 | kfree(tg->se); | |
6f505b16 PZ |
9645 | } |
9646 | ||
ec7dc8ac DG |
9647 | static |
9648 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9649 | { |
29f59db3 | 9650 | struct cfs_rq *cfs_rq; |
eab17229 | 9651 | struct sched_entity *se; |
9b5b7751 | 9652 | struct rq *rq; |
29f59db3 SV |
9653 | int i; |
9654 | ||
434d53b0 | 9655 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9656 | if (!tg->cfs_rq) |
9657 | goto err; | |
434d53b0 | 9658 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9659 | if (!tg->se) |
9660 | goto err; | |
052f1dc7 PZ |
9661 | |
9662 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9663 | |
9664 | for_each_possible_cpu(i) { | |
9b5b7751 | 9665 | rq = cpu_rq(i); |
29f59db3 | 9666 | |
eab17229 LZ |
9667 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9668 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9669 | if (!cfs_rq) |
9670 | goto err; | |
9671 | ||
eab17229 LZ |
9672 | se = kzalloc_node(sizeof(struct sched_entity), |
9673 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9674 | if (!se) |
9675 | goto err; | |
9676 | ||
eab17229 | 9677 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9678 | } |
9679 | ||
9680 | return 1; | |
9681 | ||
9682 | err: | |
9683 | return 0; | |
9684 | } | |
9685 | ||
9686 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9687 | { | |
9688 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9689 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9690 | } | |
9691 | ||
9692 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9693 | { | |
9694 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9695 | } | |
6d6bc0ad | 9696 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9697 | static inline void free_fair_sched_group(struct task_group *tg) |
9698 | { | |
9699 | } | |
9700 | ||
ec7dc8ac DG |
9701 | static inline |
9702 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9703 | { |
9704 | return 1; | |
9705 | } | |
9706 | ||
9707 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9708 | { | |
9709 | } | |
9710 | ||
9711 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9712 | { | |
9713 | } | |
6d6bc0ad | 9714 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9715 | |
9716 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9717 | static void free_rt_sched_group(struct task_group *tg) |
9718 | { | |
9719 | int i; | |
9720 | ||
d0b27fa7 PZ |
9721 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9722 | ||
bccbe08a PZ |
9723 | for_each_possible_cpu(i) { |
9724 | if (tg->rt_rq) | |
9725 | kfree(tg->rt_rq[i]); | |
9726 | if (tg->rt_se) | |
9727 | kfree(tg->rt_se[i]); | |
9728 | } | |
9729 | ||
9730 | kfree(tg->rt_rq); | |
9731 | kfree(tg->rt_se); | |
9732 | } | |
9733 | ||
ec7dc8ac DG |
9734 | static |
9735 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9736 | { |
9737 | struct rt_rq *rt_rq; | |
eab17229 | 9738 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9739 | struct rq *rq; |
9740 | int i; | |
9741 | ||
434d53b0 | 9742 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9743 | if (!tg->rt_rq) |
9744 | goto err; | |
434d53b0 | 9745 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9746 | if (!tg->rt_se) |
9747 | goto err; | |
9748 | ||
d0b27fa7 PZ |
9749 | init_rt_bandwidth(&tg->rt_bandwidth, |
9750 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9751 | |
9752 | for_each_possible_cpu(i) { | |
9753 | rq = cpu_rq(i); | |
9754 | ||
eab17229 LZ |
9755 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9756 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9757 | if (!rt_rq) |
9758 | goto err; | |
29f59db3 | 9759 | |
eab17229 LZ |
9760 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9761 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9762 | if (!rt_se) |
9763 | goto err; | |
29f59db3 | 9764 | |
eab17229 | 9765 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9766 | } |
9767 | ||
bccbe08a PZ |
9768 | return 1; |
9769 | ||
9770 | err: | |
9771 | return 0; | |
9772 | } | |
9773 | ||
9774 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9775 | { | |
9776 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9777 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9778 | } | |
9779 | ||
9780 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9781 | { | |
9782 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9783 | } | |
6d6bc0ad | 9784 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9785 | static inline void free_rt_sched_group(struct task_group *tg) |
9786 | { | |
9787 | } | |
9788 | ||
ec7dc8ac DG |
9789 | static inline |
9790 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9791 | { |
9792 | return 1; | |
9793 | } | |
9794 | ||
9795 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9796 | { | |
9797 | } | |
9798 | ||
9799 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9800 | { | |
9801 | } | |
6d6bc0ad | 9802 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9803 | |
d0b27fa7 | 9804 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9805 | static void free_sched_group(struct task_group *tg) |
9806 | { | |
9807 | free_fair_sched_group(tg); | |
9808 | free_rt_sched_group(tg); | |
9809 | kfree(tg); | |
9810 | } | |
9811 | ||
9812 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9813 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9814 | { |
9815 | struct task_group *tg; | |
9816 | unsigned long flags; | |
9817 | int i; | |
9818 | ||
9819 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9820 | if (!tg) | |
9821 | return ERR_PTR(-ENOMEM); | |
9822 | ||
ec7dc8ac | 9823 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9824 | goto err; |
9825 | ||
ec7dc8ac | 9826 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9827 | goto err; |
9828 | ||
8ed36996 | 9829 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9830 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9831 | register_fair_sched_group(tg, i); |
9832 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9833 | } |
6f505b16 | 9834 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9835 | |
9836 | WARN_ON(!parent); /* root should already exist */ | |
9837 | ||
9838 | tg->parent = parent; | |
f473aa5e | 9839 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9840 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9841 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9842 | |
9b5b7751 | 9843 | return tg; |
29f59db3 SV |
9844 | |
9845 | err: | |
6f505b16 | 9846 | free_sched_group(tg); |
29f59db3 SV |
9847 | return ERR_PTR(-ENOMEM); |
9848 | } | |
9849 | ||
9b5b7751 | 9850 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9851 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9852 | { |
29f59db3 | 9853 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9854 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9855 | } |
9856 | ||
9b5b7751 | 9857 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9858 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9859 | { |
8ed36996 | 9860 | unsigned long flags; |
9b5b7751 | 9861 | int i; |
29f59db3 | 9862 | |
8ed36996 | 9863 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9864 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9865 | unregister_fair_sched_group(tg, i); |
9866 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9867 | } |
6f505b16 | 9868 | list_del_rcu(&tg->list); |
f473aa5e | 9869 | list_del_rcu(&tg->siblings); |
8ed36996 | 9870 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9871 | |
9b5b7751 | 9872 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9873 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9874 | } |
9875 | ||
9b5b7751 | 9876 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9877 | * The caller of this function should have put the task in its new group |
9878 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9879 | * reflect its new group. | |
9b5b7751 SV |
9880 | */ |
9881 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9882 | { |
9883 | int on_rq, running; | |
9884 | unsigned long flags; | |
9885 | struct rq *rq; | |
9886 | ||
9887 | rq = task_rq_lock(tsk, &flags); | |
9888 | ||
29f59db3 SV |
9889 | update_rq_clock(rq); |
9890 | ||
051a1d1a | 9891 | running = task_current(rq, tsk); |
29f59db3 SV |
9892 | on_rq = tsk->se.on_rq; |
9893 | ||
0e1f3483 | 9894 | if (on_rq) |
29f59db3 | 9895 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9896 | if (unlikely(running)) |
9897 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9898 | |
6f505b16 | 9899 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9900 | |
810b3817 PZ |
9901 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9902 | if (tsk->sched_class->moved_group) | |
9903 | tsk->sched_class->moved_group(tsk); | |
9904 | #endif | |
9905 | ||
0e1f3483 HS |
9906 | if (unlikely(running)) |
9907 | tsk->sched_class->set_curr_task(rq); | |
9908 | if (on_rq) | |
7074badb | 9909 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9910 | |
29f59db3 SV |
9911 | task_rq_unlock(rq, &flags); |
9912 | } | |
6d6bc0ad | 9913 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9914 | |
052f1dc7 | 9915 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9916 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9917 | { |
9918 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9919 | int on_rq; |
9920 | ||
29f59db3 | 9921 | on_rq = se->on_rq; |
62fb1851 | 9922 | if (on_rq) |
29f59db3 SV |
9923 | dequeue_entity(cfs_rq, se, 0); |
9924 | ||
9925 | se->load.weight = shares; | |
e05510d0 | 9926 | se->load.inv_weight = 0; |
29f59db3 | 9927 | |
62fb1851 | 9928 | if (on_rq) |
29f59db3 | 9929 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9930 | } |
62fb1851 | 9931 | |
c09595f6 PZ |
9932 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9933 | { | |
9934 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9935 | struct rq *rq = cfs_rq->rq; | |
9936 | unsigned long flags; | |
9937 | ||
9938 | spin_lock_irqsave(&rq->lock, flags); | |
9939 | __set_se_shares(se, shares); | |
9940 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9941 | } |
9942 | ||
8ed36996 PZ |
9943 | static DEFINE_MUTEX(shares_mutex); |
9944 | ||
4cf86d77 | 9945 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9946 | { |
9947 | int i; | |
8ed36996 | 9948 | unsigned long flags; |
c61935fd | 9949 | |
ec7dc8ac DG |
9950 | /* |
9951 | * We can't change the weight of the root cgroup. | |
9952 | */ | |
9953 | if (!tg->se[0]) | |
9954 | return -EINVAL; | |
9955 | ||
18d95a28 PZ |
9956 | if (shares < MIN_SHARES) |
9957 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9958 | else if (shares > MAX_SHARES) |
9959 | shares = MAX_SHARES; | |
62fb1851 | 9960 | |
8ed36996 | 9961 | mutex_lock(&shares_mutex); |
9b5b7751 | 9962 | if (tg->shares == shares) |
5cb350ba | 9963 | goto done; |
29f59db3 | 9964 | |
8ed36996 | 9965 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9966 | for_each_possible_cpu(i) |
9967 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9968 | list_del_rcu(&tg->siblings); |
8ed36996 | 9969 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9970 | |
9971 | /* wait for any ongoing reference to this group to finish */ | |
9972 | synchronize_sched(); | |
9973 | ||
9974 | /* | |
9975 | * Now we are free to modify the group's share on each cpu | |
9976 | * w/o tripping rebalance_share or load_balance_fair. | |
9977 | */ | |
9b5b7751 | 9978 | tg->shares = shares; |
c09595f6 PZ |
9979 | for_each_possible_cpu(i) { |
9980 | /* | |
9981 | * force a rebalance | |
9982 | */ | |
9983 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9984 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9985 | } |
29f59db3 | 9986 | |
6b2d7700 SV |
9987 | /* |
9988 | * Enable load balance activity on this group, by inserting it back on | |
9989 | * each cpu's rq->leaf_cfs_rq_list. | |
9990 | */ | |
8ed36996 | 9991 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9992 | for_each_possible_cpu(i) |
9993 | register_fair_sched_group(tg, i); | |
f473aa5e | 9994 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9995 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9996 | done: |
8ed36996 | 9997 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9998 | return 0; |
29f59db3 SV |
9999 | } |
10000 | ||
5cb350ba DG |
10001 | unsigned long sched_group_shares(struct task_group *tg) |
10002 | { | |
10003 | return tg->shares; | |
10004 | } | |
052f1dc7 | 10005 | #endif |
5cb350ba | 10006 | |
052f1dc7 | 10007 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10008 | /* |
9f0c1e56 | 10009 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 10010 | */ |
9f0c1e56 PZ |
10011 | static DEFINE_MUTEX(rt_constraints_mutex); |
10012 | ||
10013 | static unsigned long to_ratio(u64 period, u64 runtime) | |
10014 | { | |
10015 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 10016 | return 1ULL << 20; |
9f0c1e56 | 10017 | |
9a7e0b18 | 10018 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
10019 | } |
10020 | ||
9a7e0b18 PZ |
10021 | /* Must be called with tasklist_lock held */ |
10022 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 10023 | { |
9a7e0b18 | 10024 | struct task_struct *g, *p; |
b40b2e8e | 10025 | |
9a7e0b18 PZ |
10026 | do_each_thread(g, p) { |
10027 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
10028 | return 1; | |
10029 | } while_each_thread(g, p); | |
b40b2e8e | 10030 | |
9a7e0b18 PZ |
10031 | return 0; |
10032 | } | |
b40b2e8e | 10033 | |
9a7e0b18 PZ |
10034 | struct rt_schedulable_data { |
10035 | struct task_group *tg; | |
10036 | u64 rt_period; | |
10037 | u64 rt_runtime; | |
10038 | }; | |
b40b2e8e | 10039 | |
9a7e0b18 PZ |
10040 | static int tg_schedulable(struct task_group *tg, void *data) |
10041 | { | |
10042 | struct rt_schedulable_data *d = data; | |
10043 | struct task_group *child; | |
10044 | unsigned long total, sum = 0; | |
10045 | u64 period, runtime; | |
b40b2e8e | 10046 | |
9a7e0b18 PZ |
10047 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
10048 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 10049 | |
9a7e0b18 PZ |
10050 | if (tg == d->tg) { |
10051 | period = d->rt_period; | |
10052 | runtime = d->rt_runtime; | |
b40b2e8e | 10053 | } |
b40b2e8e | 10054 | |
98a4826b PZ |
10055 | #ifdef CONFIG_USER_SCHED |
10056 | if (tg == &root_task_group) { | |
10057 | period = global_rt_period(); | |
10058 | runtime = global_rt_runtime(); | |
10059 | } | |
10060 | #endif | |
10061 | ||
4653f803 PZ |
10062 | /* |
10063 | * Cannot have more runtime than the period. | |
10064 | */ | |
10065 | if (runtime > period && runtime != RUNTIME_INF) | |
10066 | return -EINVAL; | |
6f505b16 | 10067 | |
4653f803 PZ |
10068 | /* |
10069 | * Ensure we don't starve existing RT tasks. | |
10070 | */ | |
9a7e0b18 PZ |
10071 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10072 | return -EBUSY; | |
6f505b16 | 10073 | |
9a7e0b18 | 10074 | total = to_ratio(period, runtime); |
6f505b16 | 10075 | |
4653f803 PZ |
10076 | /* |
10077 | * Nobody can have more than the global setting allows. | |
10078 | */ | |
10079 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10080 | return -EINVAL; | |
6f505b16 | 10081 | |
4653f803 PZ |
10082 | /* |
10083 | * The sum of our children's runtime should not exceed our own. | |
10084 | */ | |
9a7e0b18 PZ |
10085 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10086 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10087 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10088 | |
9a7e0b18 PZ |
10089 | if (child == d->tg) { |
10090 | period = d->rt_period; | |
10091 | runtime = d->rt_runtime; | |
10092 | } | |
6f505b16 | 10093 | |
9a7e0b18 | 10094 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10095 | } |
6f505b16 | 10096 | |
9a7e0b18 PZ |
10097 | if (sum > total) |
10098 | return -EINVAL; | |
10099 | ||
10100 | return 0; | |
6f505b16 PZ |
10101 | } |
10102 | ||
9a7e0b18 | 10103 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10104 | { |
9a7e0b18 PZ |
10105 | struct rt_schedulable_data data = { |
10106 | .tg = tg, | |
10107 | .rt_period = period, | |
10108 | .rt_runtime = runtime, | |
10109 | }; | |
10110 | ||
10111 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10112 | } |
10113 | ||
d0b27fa7 PZ |
10114 | static int tg_set_bandwidth(struct task_group *tg, |
10115 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10116 | { |
ac086bc2 | 10117 | int i, err = 0; |
9f0c1e56 | 10118 | |
9f0c1e56 | 10119 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10120 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10121 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10122 | if (err) | |
9f0c1e56 | 10123 | goto unlock; |
ac086bc2 PZ |
10124 | |
10125 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10126 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10127 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10128 | |
10129 | for_each_possible_cpu(i) { | |
10130 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10131 | ||
10132 | spin_lock(&rt_rq->rt_runtime_lock); | |
10133 | rt_rq->rt_runtime = rt_runtime; | |
10134 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10135 | } | |
10136 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10137 | unlock: |
521f1a24 | 10138 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10139 | mutex_unlock(&rt_constraints_mutex); |
10140 | ||
10141 | return err; | |
6f505b16 PZ |
10142 | } |
10143 | ||
d0b27fa7 PZ |
10144 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10145 | { | |
10146 | u64 rt_runtime, rt_period; | |
10147 | ||
10148 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10149 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10150 | if (rt_runtime_us < 0) | |
10151 | rt_runtime = RUNTIME_INF; | |
10152 | ||
10153 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10154 | } | |
10155 | ||
9f0c1e56 PZ |
10156 | long sched_group_rt_runtime(struct task_group *tg) |
10157 | { | |
10158 | u64 rt_runtime_us; | |
10159 | ||
d0b27fa7 | 10160 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10161 | return -1; |
10162 | ||
d0b27fa7 | 10163 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10164 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10165 | return rt_runtime_us; | |
10166 | } | |
d0b27fa7 PZ |
10167 | |
10168 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10169 | { | |
10170 | u64 rt_runtime, rt_period; | |
10171 | ||
10172 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10173 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10174 | ||
619b0488 R |
10175 | if (rt_period == 0) |
10176 | return -EINVAL; | |
10177 | ||
d0b27fa7 PZ |
10178 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10179 | } | |
10180 | ||
10181 | long sched_group_rt_period(struct task_group *tg) | |
10182 | { | |
10183 | u64 rt_period_us; | |
10184 | ||
10185 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10186 | do_div(rt_period_us, NSEC_PER_USEC); | |
10187 | return rt_period_us; | |
10188 | } | |
10189 | ||
10190 | static int sched_rt_global_constraints(void) | |
10191 | { | |
4653f803 | 10192 | u64 runtime, period; |
d0b27fa7 PZ |
10193 | int ret = 0; |
10194 | ||
ec5d4989 HS |
10195 | if (sysctl_sched_rt_period <= 0) |
10196 | return -EINVAL; | |
10197 | ||
4653f803 PZ |
10198 | runtime = global_rt_runtime(); |
10199 | period = global_rt_period(); | |
10200 | ||
10201 | /* | |
10202 | * Sanity check on the sysctl variables. | |
10203 | */ | |
10204 | if (runtime > period && runtime != RUNTIME_INF) | |
10205 | return -EINVAL; | |
10b612f4 | 10206 | |
d0b27fa7 | 10207 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10208 | read_lock(&tasklist_lock); |
4653f803 | 10209 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10210 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10211 | mutex_unlock(&rt_constraints_mutex); |
10212 | ||
10213 | return ret; | |
10214 | } | |
54e99124 DG |
10215 | |
10216 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10217 | { | |
10218 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10219 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10220 | return 0; | |
10221 | ||
10222 | return 1; | |
10223 | } | |
10224 | ||
6d6bc0ad | 10225 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10226 | static int sched_rt_global_constraints(void) |
10227 | { | |
ac086bc2 PZ |
10228 | unsigned long flags; |
10229 | int i; | |
10230 | ||
ec5d4989 HS |
10231 | if (sysctl_sched_rt_period <= 0) |
10232 | return -EINVAL; | |
10233 | ||
60aa605d PZ |
10234 | /* |
10235 | * There's always some RT tasks in the root group | |
10236 | * -- migration, kstopmachine etc.. | |
10237 | */ | |
10238 | if (sysctl_sched_rt_runtime == 0) | |
10239 | return -EBUSY; | |
10240 | ||
ac086bc2 PZ |
10241 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10242 | for_each_possible_cpu(i) { | |
10243 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10244 | ||
10245 | spin_lock(&rt_rq->rt_runtime_lock); | |
10246 | rt_rq->rt_runtime = global_rt_runtime(); | |
10247 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10248 | } | |
10249 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10250 | ||
d0b27fa7 PZ |
10251 | return 0; |
10252 | } | |
6d6bc0ad | 10253 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10254 | |
10255 | int sched_rt_handler(struct ctl_table *table, int write, | |
10256 | struct file *filp, void __user *buffer, size_t *lenp, | |
10257 | loff_t *ppos) | |
10258 | { | |
10259 | int ret; | |
10260 | int old_period, old_runtime; | |
10261 | static DEFINE_MUTEX(mutex); | |
10262 | ||
10263 | mutex_lock(&mutex); | |
10264 | old_period = sysctl_sched_rt_period; | |
10265 | old_runtime = sysctl_sched_rt_runtime; | |
10266 | ||
10267 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10268 | ||
10269 | if (!ret && write) { | |
10270 | ret = sched_rt_global_constraints(); | |
10271 | if (ret) { | |
10272 | sysctl_sched_rt_period = old_period; | |
10273 | sysctl_sched_rt_runtime = old_runtime; | |
10274 | } else { | |
10275 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10276 | def_rt_bandwidth.rt_period = | |
10277 | ns_to_ktime(global_rt_period()); | |
10278 | } | |
10279 | } | |
10280 | mutex_unlock(&mutex); | |
10281 | ||
10282 | return ret; | |
10283 | } | |
68318b8e | 10284 | |
052f1dc7 | 10285 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10286 | |
10287 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10288 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10289 | { |
2b01dfe3 PM |
10290 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10291 | struct task_group, css); | |
68318b8e SV |
10292 | } |
10293 | ||
10294 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10295 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10296 | { |
ec7dc8ac | 10297 | struct task_group *tg, *parent; |
68318b8e | 10298 | |
2b01dfe3 | 10299 | if (!cgrp->parent) { |
68318b8e | 10300 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10301 | return &init_task_group.css; |
10302 | } | |
10303 | ||
ec7dc8ac DG |
10304 | parent = cgroup_tg(cgrp->parent); |
10305 | tg = sched_create_group(parent); | |
68318b8e SV |
10306 | if (IS_ERR(tg)) |
10307 | return ERR_PTR(-ENOMEM); | |
10308 | ||
68318b8e SV |
10309 | return &tg->css; |
10310 | } | |
10311 | ||
41a2d6cf IM |
10312 | static void |
10313 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10314 | { |
2b01dfe3 | 10315 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10316 | |
10317 | sched_destroy_group(tg); | |
10318 | } | |
10319 | ||
41a2d6cf IM |
10320 | static int |
10321 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10322 | struct task_struct *tsk) | |
68318b8e | 10323 | { |
b68aa230 | 10324 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10325 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10326 | return -EINVAL; |
10327 | #else | |
68318b8e SV |
10328 | /* We don't support RT-tasks being in separate groups */ |
10329 | if (tsk->sched_class != &fair_sched_class) | |
10330 | return -EINVAL; | |
b68aa230 | 10331 | #endif |
68318b8e SV |
10332 | |
10333 | return 0; | |
10334 | } | |
10335 | ||
10336 | static void | |
2b01dfe3 | 10337 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10338 | struct cgroup *old_cont, struct task_struct *tsk) |
10339 | { | |
10340 | sched_move_task(tsk); | |
10341 | } | |
10342 | ||
052f1dc7 | 10343 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10344 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10345 | u64 shareval) |
68318b8e | 10346 | { |
2b01dfe3 | 10347 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10348 | } |
10349 | ||
f4c753b7 | 10350 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10351 | { |
2b01dfe3 | 10352 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10353 | |
10354 | return (u64) tg->shares; | |
10355 | } | |
6d6bc0ad | 10356 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10357 | |
052f1dc7 | 10358 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10359 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10360 | s64 val) |
6f505b16 | 10361 | { |
06ecb27c | 10362 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10363 | } |
10364 | ||
06ecb27c | 10365 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10366 | { |
06ecb27c | 10367 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10368 | } |
d0b27fa7 PZ |
10369 | |
10370 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10371 | u64 rt_period_us) | |
10372 | { | |
10373 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10374 | } | |
10375 | ||
10376 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10377 | { | |
10378 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10379 | } | |
6d6bc0ad | 10380 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10381 | |
fe5c7cc2 | 10382 | static struct cftype cpu_files[] = { |
052f1dc7 | 10383 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10384 | { |
10385 | .name = "shares", | |
f4c753b7 PM |
10386 | .read_u64 = cpu_shares_read_u64, |
10387 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10388 | }, |
052f1dc7 PZ |
10389 | #endif |
10390 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10391 | { |
9f0c1e56 | 10392 | .name = "rt_runtime_us", |
06ecb27c PM |
10393 | .read_s64 = cpu_rt_runtime_read, |
10394 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10395 | }, |
d0b27fa7 PZ |
10396 | { |
10397 | .name = "rt_period_us", | |
f4c753b7 PM |
10398 | .read_u64 = cpu_rt_period_read_uint, |
10399 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10400 | }, |
052f1dc7 | 10401 | #endif |
68318b8e SV |
10402 | }; |
10403 | ||
10404 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10405 | { | |
fe5c7cc2 | 10406 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10407 | } |
10408 | ||
10409 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10410 | .name = "cpu", |
10411 | .create = cpu_cgroup_create, | |
10412 | .destroy = cpu_cgroup_destroy, | |
10413 | .can_attach = cpu_cgroup_can_attach, | |
10414 | .attach = cpu_cgroup_attach, | |
10415 | .populate = cpu_cgroup_populate, | |
10416 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10417 | .early_init = 1, |
10418 | }; | |
10419 | ||
052f1dc7 | 10420 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10421 | |
10422 | #ifdef CONFIG_CGROUP_CPUACCT | |
10423 | ||
10424 | /* | |
10425 | * CPU accounting code for task groups. | |
10426 | * | |
10427 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10428 | * (balbir@in.ibm.com). | |
10429 | */ | |
10430 | ||
934352f2 | 10431 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10432 | struct cpuacct { |
10433 | struct cgroup_subsys_state css; | |
10434 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10435 | u64 *cpuusage; | |
ef12fefa | 10436 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10437 | struct cpuacct *parent; |
d842de87 SV |
10438 | }; |
10439 | ||
10440 | struct cgroup_subsys cpuacct_subsys; | |
10441 | ||
10442 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10443 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10444 | { |
32cd756a | 10445 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10446 | struct cpuacct, css); |
10447 | } | |
10448 | ||
10449 | /* return cpu accounting group to which this task belongs */ | |
10450 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10451 | { | |
10452 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10453 | struct cpuacct, css); | |
10454 | } | |
10455 | ||
10456 | /* create a new cpu accounting group */ | |
10457 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10458 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10459 | { |
10460 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10461 | int i; |
d842de87 SV |
10462 | |
10463 | if (!ca) | |
ef12fefa | 10464 | goto out; |
d842de87 SV |
10465 | |
10466 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10467 | if (!ca->cpuusage) |
10468 | goto out_free_ca; | |
10469 | ||
10470 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10471 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10472 | goto out_free_counters; | |
d842de87 | 10473 | |
934352f2 BR |
10474 | if (cgrp->parent) |
10475 | ca->parent = cgroup_ca(cgrp->parent); | |
10476 | ||
d842de87 | 10477 | return &ca->css; |
ef12fefa BR |
10478 | |
10479 | out_free_counters: | |
10480 | while (--i >= 0) | |
10481 | percpu_counter_destroy(&ca->cpustat[i]); | |
10482 | free_percpu(ca->cpuusage); | |
10483 | out_free_ca: | |
10484 | kfree(ca); | |
10485 | out: | |
10486 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10487 | } |
10488 | ||
10489 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10490 | static void |
32cd756a | 10491 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10492 | { |
32cd756a | 10493 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10494 | int i; |
d842de87 | 10495 | |
ef12fefa BR |
10496 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10497 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10498 | free_percpu(ca->cpuusage); |
10499 | kfree(ca); | |
10500 | } | |
10501 | ||
720f5498 KC |
10502 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10503 | { | |
b36128c8 | 10504 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10505 | u64 data; |
10506 | ||
10507 | #ifndef CONFIG_64BIT | |
10508 | /* | |
10509 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10510 | */ | |
10511 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10512 | data = *cpuusage; | |
10513 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10514 | #else | |
10515 | data = *cpuusage; | |
10516 | #endif | |
10517 | ||
10518 | return data; | |
10519 | } | |
10520 | ||
10521 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10522 | { | |
b36128c8 | 10523 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10524 | |
10525 | #ifndef CONFIG_64BIT | |
10526 | /* | |
10527 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10528 | */ | |
10529 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10530 | *cpuusage = val; | |
10531 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10532 | #else | |
10533 | *cpuusage = val; | |
10534 | #endif | |
10535 | } | |
10536 | ||
d842de87 | 10537 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10538 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10539 | { |
32cd756a | 10540 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10541 | u64 totalcpuusage = 0; |
10542 | int i; | |
10543 | ||
720f5498 KC |
10544 | for_each_present_cpu(i) |
10545 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10546 | |
10547 | return totalcpuusage; | |
10548 | } | |
10549 | ||
0297b803 DG |
10550 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10551 | u64 reset) | |
10552 | { | |
10553 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10554 | int err = 0; | |
10555 | int i; | |
10556 | ||
10557 | if (reset) { | |
10558 | err = -EINVAL; | |
10559 | goto out; | |
10560 | } | |
10561 | ||
720f5498 KC |
10562 | for_each_present_cpu(i) |
10563 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10564 | |
0297b803 DG |
10565 | out: |
10566 | return err; | |
10567 | } | |
10568 | ||
e9515c3c KC |
10569 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10570 | struct seq_file *m) | |
10571 | { | |
10572 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10573 | u64 percpu; | |
10574 | int i; | |
10575 | ||
10576 | for_each_present_cpu(i) { | |
10577 | percpu = cpuacct_cpuusage_read(ca, i); | |
10578 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10579 | } | |
10580 | seq_printf(m, "\n"); | |
10581 | return 0; | |
10582 | } | |
10583 | ||
ef12fefa BR |
10584 | static const char *cpuacct_stat_desc[] = { |
10585 | [CPUACCT_STAT_USER] = "user", | |
10586 | [CPUACCT_STAT_SYSTEM] = "system", | |
10587 | }; | |
10588 | ||
10589 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10590 | struct cgroup_map_cb *cb) | |
10591 | { | |
10592 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10593 | int i; | |
10594 | ||
10595 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10596 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10597 | val = cputime64_to_clock_t(val); | |
10598 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10599 | } | |
10600 | return 0; | |
10601 | } | |
10602 | ||
d842de87 SV |
10603 | static struct cftype files[] = { |
10604 | { | |
10605 | .name = "usage", | |
f4c753b7 PM |
10606 | .read_u64 = cpuusage_read, |
10607 | .write_u64 = cpuusage_write, | |
d842de87 | 10608 | }, |
e9515c3c KC |
10609 | { |
10610 | .name = "usage_percpu", | |
10611 | .read_seq_string = cpuacct_percpu_seq_read, | |
10612 | }, | |
ef12fefa BR |
10613 | { |
10614 | .name = "stat", | |
10615 | .read_map = cpuacct_stats_show, | |
10616 | }, | |
d842de87 SV |
10617 | }; |
10618 | ||
32cd756a | 10619 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10620 | { |
32cd756a | 10621 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10622 | } |
10623 | ||
10624 | /* | |
10625 | * charge this task's execution time to its accounting group. | |
10626 | * | |
10627 | * called with rq->lock held. | |
10628 | */ | |
10629 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10630 | { | |
10631 | struct cpuacct *ca; | |
934352f2 | 10632 | int cpu; |
d842de87 | 10633 | |
c40c6f85 | 10634 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10635 | return; |
10636 | ||
934352f2 | 10637 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10638 | |
10639 | rcu_read_lock(); | |
10640 | ||
d842de87 | 10641 | ca = task_ca(tsk); |
d842de87 | 10642 | |
934352f2 | 10643 | for (; ca; ca = ca->parent) { |
b36128c8 | 10644 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10645 | *cpuusage += cputime; |
10646 | } | |
a18b83b7 BR |
10647 | |
10648 | rcu_read_unlock(); | |
d842de87 SV |
10649 | } |
10650 | ||
ef12fefa BR |
10651 | /* |
10652 | * Charge the system/user time to the task's accounting group. | |
10653 | */ | |
10654 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10655 | enum cpuacct_stat_index idx, cputime_t val) | |
10656 | { | |
10657 | struct cpuacct *ca; | |
10658 | ||
10659 | if (unlikely(!cpuacct_subsys.active)) | |
10660 | return; | |
10661 | ||
10662 | rcu_read_lock(); | |
10663 | ca = task_ca(tsk); | |
10664 | ||
10665 | do { | |
10666 | percpu_counter_add(&ca->cpustat[idx], val); | |
10667 | ca = ca->parent; | |
10668 | } while (ca); | |
10669 | rcu_read_unlock(); | |
10670 | } | |
10671 | ||
d842de87 SV |
10672 | struct cgroup_subsys cpuacct_subsys = { |
10673 | .name = "cpuacct", | |
10674 | .create = cpuacct_create, | |
10675 | .destroy = cpuacct_destroy, | |
10676 | .populate = cpuacct_populate, | |
10677 | .subsys_id = cpuacct_subsys_id, | |
10678 | }; | |
10679 | #endif /* CONFIG_CGROUP_CPUACCT */ |